fix(workspace): revalidate served pages and static files on reload

The workspace pages and every file under /prompt_manager/js and
/prompt_manager/lib went out with no Cache-Control and no validator.
ComfyUI's own cache middleware only fills a missing Cache-Control for
paths ending in .js or .css, so the workspace's .mjs modules got no
header at all, and after an update the browser could keep running old
modules until a hard refresh.

Every page and static file now carries Cache-Control: no-cache and a
weak ETag over its bytes, and a request that already holds that version
gets a 304 without a body. A normal reload picks up new code and stays
cheap when nothing changed.

Closes #478

.github/workflows/viewerjs-check.yml

@@ -36,6 +36,11 @@ concurrency:
 # 12m53s, node units and the viewer suite before 11.8m of workflows), so coverage is the tight one
 # in its 18 minutes. This holds only when the npm cache hits: a cold cache adds up to 5m to npm ci,
 # and then the job limit can still end the job before the step limit does.
+# #473: firefox/webkit shard 1/2 reached its 18-minute step limit (queue run 40866252975: webkit 622 tests
+# passed in 17m56s, 48.5 worker-minutes vs 28.0 in shard 2/2; recent shard 1/2 jobs 13.7m-18.9m, shard 2/2
+# 7.3m-12.2m), because --shard splits by test count and the slow concept specs sort into shard 1. The page
+# workflows now run in three shards per browser with the limits unchanged; the per-shard times are
+# re-measured after three green main runs.
 jobs:
   frontend-coverage:
     name: frontend coverage (shard ${{ matrix.shard }}/3)
@@ -175,7 +180,7 @@ jobs:
           retention-days: 14
 
   browser-workflows:
-    name: page workflows (${{ matrix.browser }}, shard ${{ matrix.shard }}/2)
+    name: page workflows (${{ matrix.browser }}, shard ${{ matrix.shard }}/3)
     # Merge queue, main and manual runs only (#420); pull requests run the Chromium/touch coverage shards.
     if: github.event_name != 'pull_request'
     runs-on: ubuntu-latest
@@ -194,7 +199,7 @@ jobs:
       fail-fast: false
       matrix:
         browser: [firefox, webkit]
-        shard: [1, 2]
+        shard: [1, 2, 3]
     steps:
       - uses: actions/checkout@v7
       # The image's own Python 3.12 and Node 24: setup-python/setup-node install the runner's toolcache
@@ -221,9 +226,9 @@ jobs:
           path: ~/.npm
           key: npm-${{ runner.os }}-playwright-v1.64.1-${{ hashFiles('tools/vendor/package-lock.json') }}
       - run: npm ci --prefix tools/vendor --ignore-scripts
-      - name: Page workflows in ${{ matrix.browser }}, shard ${{ matrix.shard }}/2
+      - name: Page workflows in ${{ matrix.browser }}, shard ${{ matrix.shard }}/3
         timeout-minutes: 18
-        run: npm run --prefix tools/vendor test:workflows -- --project ${{ matrix.browser }} --shard ${{ matrix.shard }}/2
+        run: npm run --prefix tools/vendor test:workflows -- --project ${{ matrix.browser }} --shard ${{ matrix.shard }}/3
       # Traces and the blob report (test-results/blob/) of a failed, timed-out or cancelled step.
       - name: Upload browser traces
         if: always()

TESTING.md

@@ -1830,8 +1830,9 @@ Inspect plan keeps D60–D66 (plan D60 → D83, D61 → D84, D65 → D88, D68
   blocks' context ("Current file image tags" in History). An empty source draws no
   section. Repeats are removed within a source only, so a linked tag that is also an
   image tag shows in both sections and each count is true. D21's filter rules are
-  unchanged. History's current-file read carries no `image_tags` facet (D88), so there
-  only the linked section shows. `tests/browser/inspector-concept.spec.mjs` wraps the
+  unchanged. History's current-file read carries the same `image_tags` facet as the Images
+  read (#457), so History's inline current file and its viewer show the image tags section
+  too (`test_current_inspection_carries_the_files_image_tags`). `tests/browser/inspector-concept.spec.mjs` wraps the
   concept's row in the same sections at runtime (`withTagSections`, the D154 rule
   injected) and compares the measured sections (frame, heading, label, row); a
   mutation of the label color, the row margin or the top rule fails it. Browser tests:
@@ -2442,10 +2443,38 @@ Recorded deviations on this page:
   view" filters are active, and plain "Filters" otherwise; the text is its accessible
   name, so the count is spoken. The concept has no indicator, and without one a Tags
   filter set by Tags & categories Browse was invisible on the shelf. The Library counts
-  Tags; Images counts Source, Folder, Media and Tags (each once). The toolbar category is
+  Tags, Folder and Tag match (D150); Images counts Source, Folder, Media and Tags (each once). The toolbar category is
   not a Refine filter and is not counted. The caption's count is a `role="status"`
   region, written only when it changes and kept shown while the shelf reloads, so
   Load more and scrolling never re-announce it.
+- **D150** (#444, cutover #334): "Refine this view" on the Library holds the concept
+  Filters dialog's **Folder** ("All output folders", then each folder below an output
+  root as a tree built by path segment (each folder directly under its own parent,
+  case variants as distinct siblings, siblings case-insensitive then exact), indented
+  by depth, read from `GET /prompt_manager/library/options?include=folders`
+  when the dialog opens; the chosen folder stays listed even when the list lacks it) and
+  **Tags**, plus **Tag match** ("All of these tags" / "Any of these tags" / "Prompts without tags", D149) drawn as the
+  concept's own select; the concept's "Prompt evidence" and prototype note are absent.
+  `GET /prompt_manager/library` takes `folder` (`/`-joined; `\` is a separator only on
+  Windows, on POSIX it is part of a folder's name as the listing shows it; `.`, `..`,
+  empty segments and, on Windows, drives are 400) and `tag_mode` (`all` default, `any`, `untagged` from D149), and echoes
+  both; a client whose non-default value is not echoed (an un-restarted older server)
+  shows the read error, never the unfiltered shelf. A folder keeps a parent with one
+  linked image (`generated_images.prompt_id`) in it or below it under any spelling of
+  any output root (the root as configured or detected and after `realpath`, so a
+  symlinked output folder works): `file_path` (`normcase(normpath(abspath))`) is
+  compared with prefixes built by the same normalization, so case folding is the
+  platform's own, including non-ASCII on Windows. The folder is an outer AND around the
+  parent and pair arms; Any puts one `t.name IN (...)` in each arm, so a card matches
+  when the parent or one active pair carries any tag (text and category still AND
+  inside each arm); with one tag Any equals All. Filters counts Tags, Folder and a Tag
+  match other than All (D44). Saved Library views store `tag_mode` (`all|any|untagged`) and
+  `folder` (text 1024); older rows read the defaults. Tests:
+  `tests/test_library_folders.py` (pure prefixes and names with `posixpath`/`ntpath`,
+  drive, UNC, non-ASCII, symlinked root, the any-tag union), `test_api_library.py`,
+  `test_saved_views.py`, the library-state and saved-views JS units, and
+  `tests/browser/workspace-library-filters.spec.mjs` (behavior, the Refine dialog against
+  the concept Filters dialog, artifact captures). Windows: run `portability.yml` on the PR.
 
 `tests/browser/library-concept.spec.mjs` compares the shelf (#249 E3) on the
 concept-shaped fixture, which creates the concept's prompts last to first so Recently
@@ -2811,10 +2840,12 @@ mapped differences"):
 - **D60** (heading action, LI-3): "Open file metadata" (`ca-primary`) replaces the
   concept's "Open sample metadata". It is `aria-disabled="true"` and
   `aria-describedby="local-file-note"` (the drop note) until a classified selection is
-  ready, then opens the shared viewer on that file.
+  ready, then opens the shared viewer on that file. Amended by D155: choosing the file
+  already opens it; the action reopens it.
 - **D61** (copy, LI-3): the surface `p` reads "Read embedded metadata in this browser.
   The file is not uploaded or added to your library."; the drop note "Your file stays in
-  this browser. Its metadata opens with “Open file metadata”."; the chooser label "Choose
+  this browser. Its metadata opens with “Open file metadata”." (D155: "Your file stays in
+  this browser. Choosing it opens its metadata."); the chooser label "Choose
   an image or video for a local preview"; the toast "Local preview only. No file
   uploaded; metadata is available above.". The h2 "Inspect without importing" is the
   concept's, verbatim.
@@ -2868,6 +2899,7 @@ mapped differences"):
   by the next selection attempt or after the concept's toast interval (6.5 s). The toast
   "Local preview only…; metadata is available above." shows only for a selection whose
   status is empty or the AVI/WMV note, never beside an unreadable or failed preview.
+  Amended by D155: only the AVI/WMV note keeps the toast; an empty status opens the viewer.
 - **D64** (drop, LI-6): `dragover`/`drop` on `.file-drop` accept exactly one File, read
   only `dataTransfer.files`/`items` (never its strings) and change nothing visually.
   While the page is mounted, `window` `dragover`/`drop` call `preventDefault`, so a missed
@@ -2880,6 +2912,36 @@ mapped differences"):
   Modified with no Path row and no Show in folder, and ends with "Local file; no saved
   library association was checked." and Download original. There is no Prompt history
   action and no category or tag chips.
+- **D155** (#448, maintainer choice 2026-10-28, beyond the concept): choosing or dropping
+  a file opens the shared viewer on it at once, with `#viewer-details` showing, in one
+  step; the concept previews it and waits for its heading action. The viewer opens once
+  per selection, after the session has selected the record (a microtask after `ready`),
+  only while that selection is current and no other dialog is open. A new selection
+  that starts while the viewer shows the previous file closes it first, so no element
+  holds the old `blob:` URL when the session revokes it (`URLs and media end with their
+  owning session`); the new file's viewer opens once it is ready. A video opens the
+  viewer only once its inline preview has settled (metadata, an error, or D62's 15 s):
+  replacing a video while both the inline and the viewer's video were still starting
+  deadlocked WebKit (`a video replaced while still starting is reset once it settles`
+  hung 2 of 2 runs before, passes 5 of 5 after). It opens for a selection with no status and for one
+  without a preview URL ("The preview could not be displayed. Metadata and the original
+  remain available in the viewer."); an unreadable file stays on the page with its
+  alert, and AVI/WMV keeps its note and the toast (the viewer could not play it and
+  reads no metadata); "Open file metadata" opens either. An image or video that fails to
+  decode after the viewer opened shows the viewer's own media error; the page's alert
+  is there on close. The toast no longer shows for an opened file (the viewer is the
+  confirmation), and the drop note reads "Your file stays in this browser. Choosing it
+  opens its metadata.". On this page the viewer, by choice or by the heading action,
+  opens with the details showing even when "Collapse viewer details by default" is on
+  (D97); the toggle is still drawn then and starts as "Hide details" (`aria-expanded`
+  true). Closing returns focus to the chooser (the heading action when it opened the
+  viewer), with the preview, the chooser naming the file and the heading action in
+  place; the next file chosen or dropped replaces it and opens again. A drop is still
+  ignored while the viewer is open (D64). Nothing else changes: the file is read in this
+  browser only. Browser tests: "D155: choosing a file opens the viewer with its details
+  in one click", "D155: Inspect shows the details even when the viewer collapses them by
+  default (amends D97)", "D155: an unreadable file stays on the page with its alert" and
+  "D155: at 390px the opened details are reachable below the media".
 - **D117** (Copy settings summary, LI-4): the legacy metadata
   page's COPY ALL, consolidated into the shared full inspector (the viewer's
   `#viewer-details`, so Images, the recorded history's current file and Inspect all have
@@ -3238,7 +3300,8 @@ until this browser dismissed the 3.2.0 update notice (`pm_update_notice_dismisse
 the legacy admin page's key); Later dismisses it without scanning, and Review scan
 dismisses it as the legacy Scan does. "+ Add source" asks validate-source, then adds
 the folder to the host draft of `gallery_root_paths`; Remove takes one out; "Watch for
-new outputs" (`monitoring_enabled`) and Worker threads (`worker_threads`) are host
+new outputs" (`monitoring_enabled`), "Save prompts from other workflows"
+(`save_foreign_runs`, D158) and Worker threads (`worker_threads`) are host
 drafts too, and Save writes them with the revision. A save the server refuses because a
 scan runs toasts "Another job is running: <title>." with a link to that job (D57) and
 keeps the drafts; a save refused for one source says the server's message as the status
@@ -3487,7 +3550,7 @@ the connection's `total_changes` is 0 at close and the data reads back unchanged
 `-shm` bytes are not compared: every WAL reader rewrites its read marks there, so no
 reader can keep it byte-identical (`test_health_check_never_writes`).
 
-Recorded deviations (plan D60–D75, D79–D82 renumbered D92–D107, D109–D110 and D111–D114; D121–D134 assigned). D94,
+Recorded deviations (plan D60–D75, D79–D82 renumbered D92–D107, D109–D110 and D111–D114; D121–D134 and D158 assigned). D94,
 the per-section hand-off panel, was retired by S12 (#329): every section is native.
 
 - **D92** (shell): no "Sample configuration" chip. The Save status is truthful: "No
@@ -3558,10 +3621,41 @@ the per-section hand-off panel, was retired by S12 (#329): every section is nati
   card's title and file line, Info and Tag stay at the end of the row), "Collapse viewer
   details by default" (`sidebarCollapsedByDefault`; the viewer opens with
   `#viewer-details` hidden and a "Show details"/"Hide details" toggle with
-  `aria-expanded`, the stage taking the whole width), and the operation "Reset display
+  `aria-expanded`, the stage taking the whole width; amended by D155: on Inspect a file
+  the viewer opens with the details showing and the toggle reading "Hide details"; amended by D156: the row moved
+  to the "Viewer" surface), and the operation "Reset display
   defaults" ("Restore this browser’s display defaults, then save. Other settings are
   kept.", button "Reset"), which drafts the gallery's defaults for the browser rows only.
   The node search timeout shows the server's real value, not the sample 30. CSS under D97.
+- **D156** (#450, maintainer request 2026-10-28, beyond the concept): every viewer (the
+  shared `dialog#viewer-dialog` of Images, Library, History, collections and Inspect) has
+  `button#viewer-mode.icon-btn` in `.dialog-header`, right before ✕ (a direct child, so
+  the concept's `.dialog-header > button` and D34's coarse 40px rule cover it; 8px from
+  ✕). It names the view it switches to in its accessible name and `title` ("Full screen
+  view" / "Centered view") with a `maximize` / `minimize` outline icon, and keeps focus
+  after a switch. Full screen (`#viewer-dialog.viewer-fullscreen`) fills the browser
+  window edge to edge: `inset: 0`, no width or height cap, no margin, border or radius
+  (`.ca-tools #viewer-dialog.viewer-fullscreen[open]`, beating the concept's box, D34's
+  max-height and the ≤700px height); the layout inside is the concept's. Centered is the
+  concept's framed dialog, unchanged. In full screen the details stay toggleable: the
+  D97 "Show details"/"Hide details" toggle is in the toolbar whenever the view is full
+  screen or "Collapse viewer details by default" is on; going back to Centered without
+  that setting removes the toggle and shows the details. The view is not the browser's
+  Fullscreen API (Esc keeps one meaning, closing the viewer; WebKit has no element
+  fullscreen for a modal dialog; the workspace also runs in ComfyUI popups). The viewer
+  opens in the view System › Appearance › **Viewer** › "Open viewer in" (Centered,
+  Full screen) chooses: a browser display preference, `viewerOpenMode` (`centered` by
+  default, normalized in `gallery-settings.js`) in the shared `gallerySettings`; no
+  server setting. The button changes only the current view (navigation in the viewer
+  keeps it; the next open uses the setting). "Viewer" is its own surface after
+  "Browsing & display" ("How the image viewer opens; the button beside ✕ switches the
+  current view."), holding "Open viewer in" and the moved "Collapse viewer details by
+  default"; "Reset display defaults" still restores both. At ≤700px (the concept's
+  breakpoint, where the viewer is already 96vw × 94dvh) the button is hidden
+  (`display: none !important` against the concept's `.icon-btn` rule), popup windows
+  that narrow included; the stored view still applies there. Browser tests:
+  `tests/browser/workspace-viewer-mode.spec.mjs`; the concept comparisons add the
+  button to the concept's header at runtime and map the Viewer surface.
 - **D98** (Sources): each configured source's row has a plain Remove after its scan
   button; nothing changes until Save. A source added this session reads "<path> · not
   saved yet" with Remove only until Save; a configured source the server cannot use
@@ -3976,6 +4070,26 @@ the per-section hand-off panel, was retired by S12 (#329): every section is nati
   Cancel-only review matches the concept review shell` (system-tagging-concept.spec) and
   the batch tests of workspace-system-tagging.spec; the mutation check renames each grid
   label in turn.
+- **D158** (Sources, #247 D-Q4): the concept has no switch for prompts from workflows
+  without a PromptManager node; the maintainer chose a visible switch over a config key.
+  A `.setting-row` in the concept's toggle markup follows "Watch for new outputs":
+  "Save prompts from other workflows" (`#foreign-runs`, host key `save_foreign_runs`,
+  config.json `gallery.monitoring.save_foreign_runs`, default on), "New outputs of
+  workflows without a PromptManager node add their prompt to the library when it is
+  certain.". It is saved with the revision like the other host rows and applied at once
+  (the monitor reads it for every file), so a save names no restart. While "Watch for
+  new outputs" is drafted off, a `p.small` under the description says "Applies while
+  Watch for new outputs is on." (the switch's `aria-describedby`; the value is kept) and
+  the switch is disabled, except while a source still reads "watching until restart"
+  (D128): the running observer still delivers files then, so the value can be set. The
+  server stores no foreign prompt while `monitoring_enabled` is saved off, whatever the
+  observer does.
+  Pinned by `sources matches the concept` (the mapping inserts the row after the
+  monitoring row), `sources comparison detects mutations` (a missing row and a changed
+  label are caught), `D158: Save prompts from other workflows saves save_foreign_runs
+  live; off while monitoring is off (#247)` (keyboard on desktop, a tap on touch) and
+  `D158: saved off while the monitor still watches until a restart, the foreign row can
+  still be set (D128)` (workspace-system-sources.spec) and the system-sources node test.
 - **D109** (Logs, S12, #329): the concept's Logs surface on the live server. The toolbar
   keeps "Level", Refresh and the download (renamed "Download log": the active log,
   `/logs/download/prompt_manager.log`, through the api client), with all six minimum
@@ -5903,6 +6017,70 @@ Tests: `tests/test_file_prompt.py` (one test per ladder step and junk class),
 saver, a legacy-linked file left as it is, `_BatchHistory.recorded`) and
 `tests/test_recorded_saver_anchor.py` `TestInspectionResolution`.
 
+**In the monitor.** An output whose workflow has no positive PromptManager node
+(`_find_prompt_from_metadata` says `foreign`) used to be logged and skipped. It now
+goes through the same ladder with the `MONITOR` policy (`utils/foreign_runs.py`, the
+handler's `_ingest_foreign`, `_recheck_foreign`, `_store_foreign`):
+
+- **At settle.** The ladder runs without history. Certain (`pair` or `positive`): the
+  prompt is stored and the file linked at once. `history` (an undecided graph inside
+  ComfyUI's output folder): the file is registered in its own waiting set, then the
+  ladder runs again at once with the file's recorded output, because the run may have
+  been sealed before the file settled. When that lookup finds a record, its answer is
+  final: certain is stored, anything else is logged "No certain prompt for <file>;
+  left unlinked" and the file stops waiting (no seal is coming). When it finds none,
+  only the inspector's or the recorded saver's answer is stored; anything else (the
+  legacy floor included) keeps waiting for the re-check. Anything else at the first
+  pass is logged "No certain prompt for <file>; left unlinked".
+- **After the run ends.** When the run's outputs are sealed, each waiting foreign file
+  gets one re-check (`ForeignRecheck(reference)`) with the sealed output reference,
+  so history is asked with the spelling it recorded (a Windows folder whose case
+  differs from the disk still links). Not certain: "No certain prompt for <file>
+  after its run ended; left unlinked". An ambiguous inspector answer is never given
+  the legacy floor in the monitor (D-Q3); an untraceable one still is.
+- **Waiting sets.** `_awaiting_foreign` and `_foreign_due` sit beside the keyless sets
+  with the same bounds (1000 files, one hour) but are separate, so a large foreign run
+  never evicts a waiting keyless file. An expired entry is logged "No recorded run
+  for <file>; left unlinked", an evicted one (waiting or due) "Too many files wait for
+  their run; <file> left unlinked", both after the lock is released. A due re-check
+  is dropped only when it was scheduled, never a newer reference for the same file. `awaiting_foreign()` lists them;
+  `shutdown()` and `close()` clear them, and a closed handler re-checks nothing.
+- **Store.** One write through `survive_switch` (rerun after a Move library): a pair
+  through `ingest_resolved_pair(..., marks=RUN_MARKS)`, a positive through
+  `store_scanned_prompt_or_raise(..., RUN_MARKS)`. A created prompt carries category
+  `scanned`, tag `auto-scanned` and the note `Saved from a workflow run: <file>`; a
+  file linked to another prompt is `conflict` and keeps its link. Logged at info
+  "Foreign run: <status> prompt <id> for <file> (<source>)"; a failure is logged and
+  the file stays unlinked.
+- **Never.** No tracker FIFO entry is read or consumed, nothing is counted (run count
+  stays 0, no `last_used_at`), and nothing is queued for Needs review.
+- **The setting.** `GalleryConfig.SAVE_FOREIGN_RUNS` (`gallery.monitoring.
+  save_foreign_runs`, default true, only a bool applies) is read for every file, at
+  settle, at the re-check and again just before the store, together with
+  `MONITORING_ENABLED`: either saved off stops foreign writes at once (the running
+  observer watches until a restart, D128); off logs "No PromptManager prompt in
+  workflow, skipping image" as before. System › Sources & scans "Save prompts from other workflows" (D158)
+  saves it through `POST /prompt_manager/settings` with no restart field; the
+  diagnostics report lists it.
+
+Tests: `tests/test_run_attribution.py` `TestMonitorForeignRuns` (dataset branches
+linked after the seal, a seal before settle, a name two runs reported, the setting
+off at settle and at the re-check, a plain workflow linked at settle with no run
+count and the tracker queue untouched, the sealed reference's spelling, the separate
+bound, expiry, a seal while the pending set is full, shutdown, a failed store, no
+record at settle keeps waiting, a record without an answer ends the wait, monitoring
+saved off, the setting read before the store, a seal between registering and the
+settle lookup, a newer due reference kept, evictions logged, one link per file),
+`tests/test_image_linking.py` `TestForeignWorkflows`, `tests/test_image_monitor_singleton.py`
+(close drops a foreign re-check), `tests/test_config.py`,
+`tests/test_settings_sources.py` `TestSaveForeignRunsSetting`, and the real-host
+`tests/comfyui/image-attribution.spec.mjs` case "an output of a workflow without a
+PromptManager node creates and links its prompt; a second run reuses it" (ComfyUI's
+own `CLIPTextEncode`, `KSampler` and `SaveImage`), and "each branch of a
+shared-prefix workflow without a PromptManager node links to its own prompt after the
+run" (two branches, one save prefix: the recorded saver names each file's branch).
+The fixture route's GET also lists `foreign`, the waiting foreign files.
+
 ## Linked prompt text (#343)
 
 A link drives `text` on every PromptManager node, and `negative_text` on the Pair,
@@ -6183,7 +6361,7 @@ The versioned public reads are:
 | `GET /prompt_manager/library/{positive_id}/history/pairs/{pair_id}/newest` | The pair's newest recorded run for the history run dot (#249 D26): `{run: {run_id, anchor_id, context_id}}` or `{run: null}` |
 | `GET /prompt_manager/history/runs/{run_id}` | Frozen fields, per-field provenance, capture and outcome for a known UUID |
 | `GET /prompt_manager/history/runs/{run_id}/outputs` | Ordered host-reported occurrences, including repeated locations |
-| `GET /prompt_manager/history/runs/{run_id}/output/current/{media,thumbnail,inspection}` | Authorized current file at that reported location, using the returned epoch/scope URL |
+| `GET /prompt_manager/history/runs/{run_id}/output/current/{media,thumbnail,inspection}` | Authorized current file at that reported location, using the returned epoch/scope URL; `inspection` carries the shared inspector envelope including the file's `image_tags` facet (#457) |
 
 The newest-pair read takes no query. It answers 404 when the positive is not an
 eligible saved prompt or the pair is not one of its pairs. It walks the positive's exact
@@ -6487,15 +6665,16 @@ Deploying this backend needs a ComfyUI restart (Python and route changes).
 
 `#/tags` ("Tags & categories", page title "Tags") is a native workspace route; it no
 longer hands off to the legacy tag manager. That page stays reachable from Overview's
-"Open existing tag manager" card for tag search and sort, the untagged entry and
-multi-tag AND/OR browsing; bulk prompt delete stays on the legacy admin page (bulk tags:
-the Prompt library's selection bar, D151). A
-prompt's category is changed in the Prompt library editor, on many prompts at once with
-the Prompt library's selection bar (D148, #437), or with the legacy admin page's "Set
-Category" (both `POST /prompt_manager/bulk/category`);
+"Open existing tag manager" card for its own page; tag search and sort, the untagged
+entry (D149) and multi-tag OR (the Library's Tag match, D150) moved to the workspace
+(#444); bulk prompt delete stays on the legacy admin page (bulk tags: the Prompt
+library's selection bar, D151). A prompt's category is changed in the Prompt library
+editor, on many prompts at once with the Prompt library's selection bar (D148, #437), or
+with the legacy admin page's "Set Category" (both `POST /prompt_manager/bulk/category`);
 the Images selection bar's Set category sets images' own categories only (D140, #332),
 and Manage category renames a value on prompts, recipes and images together (D144). The screen consumes only the `/prompt_manager/vocabulary/*` routes
-above and adds no route, so PR B needs no ComfyUI restart: a browser refresh is enough.
+above and adds no route, so PR B needed no ComfyUI restart; D149 (#444) adds read
+parameters to `GET tags` and `GET /prompt_manager/library`, which do.
 
 Screen: a `+ Create tag` heading action, a Tag vocabulary table (columns Tag, Prompts,
 Images counting the files that carry the tag (`image_count`, #249 C2), and Actions with Browse and
@@ -6547,6 +6726,36 @@ Apply; "Apply" instead of "Apply in prototype"; Delete tag… inside Rename or m
 Browse sets the exact filter; singular "1 positive prompt"; the production list appends
 fixture-only tags after the concept's sample tags.
 
+- **D149** (#444, cutover #334; the concept's renderTags has neither): the Tag vocabulary
+  surface shows, after its paragraph, an "Untagged prompts" `.operation` row ("N positive
+  prompts", Browse) and then the tag search and sort, drawn as the concept page toolbar's
+  `search.toolbar` > `label.search-field` (search icon, `input#tag-search` "Search tags…",
+  maxlength 200) and a `select#tag-sort` (Oldest first, the default and the old order;
+  Name A–Z; Name Z–A; Most prompts; Fewest prompts). Search is a literal, case-insensitive
+  substring on the server (`GET /prompt_manager/vocabulary/tags?search=&sort=`; ASCII-only
+  case folding, like SQLite `LIKE`/`NOCASE`), debounced 250 ms; it resets only the tag list
+  to page one, keeps through command re-reads, reads "No tags match this search." when
+  empty and is announced in a visually hidden status ("N tags"). Every sort breaks ties by
+  ascending id; a cursor belongs to one search and sort (scope-hashed; another is 400 with
+  "refresh"); a count sort's counts can move between pages only through a write, which
+  re-reads from page one. Each tags page reports `untagged_count` (eligible Library parents
+  with no tag and no active pair carrying one, the complement of every tag's Browse; legacy
+  counted every prompt without `prompt_tags` and hid the entry at 0; this row always
+  shows) and echoes `search` and `sort`. Browse opens the Library with `tag_mode=untagged`
+  (`GET /prompt_manager/library`, echoed as `tag_mode`; with a `tag` it is 400): "Filters ·
+  1", and Refine this view adds **Tag match** after Tags ("All of these tags" / "Any of these tags" (D150) / "Prompts
+  without tags", the latter disables Tags and drops them on Apply). A client whose request
+  the server did not echo (an un-restarted older server) shows the read error with Retry,
+  never an unfiltered list. Saved Library views store `tag_mode` (`all|any|untagged`, default
+  `all`); untagged with tags is refused on write and reads back as `all`; an older build
+  opens an untagged view as the whole Library. The concept comparison holds the row to the
+  concept's first category row, its top to the concept table's offset, the table to that
+  offset plus the additions' height (`d149`) and the toolbar to the concept Library
+  toolbar (`toolbarInventoryOf`), with a mutation check. Tests:
+  `tests/test_vocabulary.py` (`TestVocabularyTagSearchAndSort`, `TestLibraryTagMode`),
+  `test_api_vocabulary.py`, `test_api_library.py`, `test_saved_views.py`, the
+  vocabulary/library/saved-views JS units and `tests/browser/workspace-vocabulary-search.spec.mjs`
+  (behavior plus artifact captures).
 - **D143** (#332): category rows count the images whose own category is the value
   (`image_count`, the stored rows, missing files included): "N positive prompts · M
   images" beside prompts, "M images" for an image-only value, which follows every

database/execution_contexts.py

@@ -7,7 +7,11 @@ from typing import Callable, ContextManager, Iterator
 
 try:
     from ..utils.execution_history_events import RunFinish, encode_command
-    from ..utils.hashing import stored_prompt_text, stored_prompt_text_sql
+    from ..utils.hashing import (
+        generate_prompt_hash,
+        stored_prompt_text,
+        stored_prompt_text_sql,
+    )
     from ..utils.execution_contexts import (
         CAPTURE_REASONS,
         ContextBundle,
@@ -16,7 +20,11 @@ try:
     )
 except ImportError:
     from utils.execution_history_events import RunFinish, encode_command
-    from utils.hashing import stored_prompt_text, stored_prompt_text_sql
+    from utils.hashing import (
+        generate_prompt_hash,
+        stored_prompt_text,
+        stored_prompt_text_sql,
+    )
     from utils.execution_contexts import (
         CAPTURE_REASONS,
         ContextBundle,
@@ -149,9 +157,12 @@ def _role_hints(role: str) -> str:
 
 # Set form of the read-time prompt-hint rule in get_execution_contexts():
 # a stored role hint is live only in the current storage epoch, and only when
-# exactly one prompt has its raw snapshot text or that text's stored form
-# (outer whitespace stripped, #438), and that is the hinted one (_prompt_hint).
-# Evaluated once per statement, never per prompt row.
+# the hinted prompt is the one match of _prompt_hint: by text (the raw
+# snapshot text or its stored form, outer whitespace stripped, #438), or, only
+# when no prompt has either text, by dedup hash (case folded too, #460).
+# The hash is a fallback, so a hint #438 kept is never made ambiguous.
+# Needs register_prompt_functions (pm_prompt_hash). The hint rows are the
+# outer loop (CROSS JOIN), so each prompts lookup is by index.
 # Defines the CTE ``live_hint_prompts(prompt_id)``; prefix it with WITH.
 LIVE_HINT_CTES = (
     "stored_hints(prompt_id, raw, stored) AS ("
@@ -160,7 +171,12 @@ LIVE_HINT_CTES = (
     + _role_hints("negative")
     + "), hint_matches(prompt_id, raw, stored, match_id) AS ("
     "SELECT h.prompt_id, h.raw, h.stored, p.id FROM stored_hints h "
-    "JOIN prompts p ON p.text COLLATE BINARY IN (h.raw, h.stored)"
+    "CROSS JOIN prompts p ON p.text COLLATE BINARY IN (h.raw, h.stored) "
+    "UNION ALL "
+    "SELECT h.prompt_id, h.raw, h.stored, p.id FROM stored_hints h "
+    "CROSS JOIN prompts p ON p.hash = pm_prompt_hash(h.raw) "
+    "WHERE NOT EXISTS (SELECT 1 FROM prompts q "
+    "WHERE q.text COLLATE BINARY IN (h.raw, h.stored))"
     "), live_hint_prompts(prompt_id) AS ("
     "SELECT prompt_id FROM hint_matches GROUP BY prompt_id, raw, stored "
     "HAVING COUNT(*) = 1 AND MIN(match_id) = prompt_id)"
@@ -169,20 +185,32 @@ LIVE_HINT_PROMPT_IDS_SQL = (
     "WITH " + LIVE_HINT_CTES + " SELECT DISTINCT prompt_id FROM live_hint_prompts"
 )
 
+_BY_TEXT = "SELECT id FROM prompts WHERE text COLLATE BINARY IN (?,?) LIMIT 2"
+
+
+def _text_matches(conn: sqlite3.Connection, raw: str) -> list:
+    return [row[0] for row in conn.execute(_BY_TEXT, (raw, stored_prompt_text(raw)))]
+
 
 def _prompt_hint(conn: sqlite3.Connection, raw: str | None) -> int | None:
-    """The one prompt whose text is ``raw`` or the stored form of it (#438).
+    """The one prompt that stores ``raw`` (#438), else the one it counts toward.
 
-    The Library stores ``text.strip()``; the exact form keeps hints of older
-    rows whose stored text kept outer whitespace.
+    By text first: ``raw`` or ``raw.strip()`` (the exact form keeps hints of
+    older rows that kept outer whitespace). Only when no prompt has either
+    text, by the dedup hash, which also folds case (#460).
     """
     if raw is None:
         return None
-    rows = conn.execute(
-        "SELECT id FROM prompts WHERE text COLLATE BINARY IN (?,?) LIMIT 2",
-        (raw, stored_prompt_text(raw)),
-    ).fetchall()
-    return rows[0][0] if len(rows) == 1 else None
+    matches = _text_matches(conn, raw)
+    if not matches:
+        matches = [
+            row[0]
+            for row in conn.execute(
+                "SELECT id FROM prompts WHERE hash=? LIMIT 2",
+                (generate_prompt_hash(raw),),
+            )
+        ]
+    return matches[0] if len(matches) == 1 else None
 
 
 def _hints(
@@ -191,24 +219,16 @@ def _hints(
     pos_id, neg_id = _prompt_hint(conn, positive), _prompt_hint(conn, negative)
     stored = None if negative is None else stored_prompt_text(negative)
     pair_id = None
+    # A known negative that names no prompt because several share its text
     negative_ambiguous = (
-        stored is not None
-        and neg_id is None
-        and conn.execute(
-            "SELECT 1 FROM prompts WHERE text COLLATE BINARY IN (?,?) LIMIT 1",
-            (negative, stored),
-        ).fetchone()
-        is not None
+        stored is not None and neg_id is None and bool(_text_matches(conn, negative))
     )
     if pos_id is not None and stored is not None and not negative_ambiguous:
         rows = conn.execute(
-            "SELECT pp.id FROM prompt_pairs pp LEFT JOIN prompts n ON "
-            "n.id=pp.negative_prompt_id "
+            "SELECT pp.id FROM prompt_pairs pp "
             "WHERE pp.positive_prompt_id=? AND ((pp.negative_prompt_id IS NULL AND "
-            "?='') "
-            "OR (pp.negative_prompt_id=? AND n.text COLLATE BINARY IN (?,?))) "
-            "LIMIT 2",
-            (pos_id, stored, neg_id, negative, stored),
+            "?='') OR pp.negative_prompt_id=?) LIMIT 2",
+            (pos_id, stored, neg_id),
         ).fetchall()
         if len(rows) == 1:
             pair_id = rows[0][0]

database/health.py

@@ -23,10 +23,10 @@ from pathlib import PurePosixPath, PureWindowsPath
 from urllib.parse import quote
 
 try:
-    from ..utils.hashing import generate_prompt_hash
+    from ..utils.hashing import register_prompt_functions
     from .models import normalize_image_path
 except ImportError:
-    from utils.hashing import generate_prompt_hash
+    from utils.hashing import register_prompt_functions
     from database.models import normalize_image_path
 
 logger = logging.getLogger("prompt_manager.database.health")
@@ -125,12 +125,8 @@ def _foreign_keys(conn, _roots):
     )
 
 
-def _prompt_hash(text):
-    return generate_prompt_hash(text) if isinstance(text, str) else None
-
-
 def _canonical_hashes(conn, _roots):
-    conn.create_function("pm_prompt_hash", 1, _prompt_hash, deterministic=True)
+    register_prompt_functions(conn)  # pm_prompt_hash, shared with history (#460)
     return _counted(
         f"Prompt {row[0]}: the stored hash does not match its text"
         for row in conn.execute(

database/history_reads.py

@@ -16,9 +16,9 @@ from .execution_outputs import _capture, _REFERENCE_COLUMNS
 from .library import _ELIGIBLE
 
 try:
-    from ..utils.hashing import stored_prompt_text, stored_prompt_text_sql
+    from ..utils.hashing import prompt_key, stored_prompt_text
 except ImportError:
-    from utils.hashing import stored_prompt_text, stored_prompt_text_sql
+    from utils.hashing import prompt_key, stored_prompt_text
 
 RunKey = tuple[int, str, int, str]
 
@@ -115,12 +115,12 @@ _MATCHING = (
 )
 
 
-def _scope_text_ids(conn: sqlite3.Connection, scope: str) -> str:
-    """Every recorded raw text whose stored form is ``scope``, as a JSON array.
+def _scope_text_ids(conn: sqlite3.Connection, key: str) -> str:
+    """Every recorded raw text whose prompt key is ``key``, as a JSON array.
 
-    The Library stores ``text.strip()`` while a run keeps its exact raw text,
-    so a raw text joins the scope of the prompt it saves (#438). The values
-    are compared, never only their digests.
+    A run keeps its exact raw text, while the Library saves and counts it as
+    the prompt with its key (outer whitespace #438, case #460). The values are
+    compared, never only their digests; ``pm_prompt_key`` is Python's key.
     """
     return json.dumps(
         [
@@ -130,16 +130,16 @@ def _scope_text_ids(conn: sqlite3.Connection, scope: str) -> str:
                 "SELECT t.text_id FROM (SELECT DISTINCT raw_text_id AS text_id "
                 "FROM execution_prompt_snapshots) s CROSS JOIN "
                 "execution_context_texts t WHERE t.text_id=s.text_id AND "
-                f"{stored_prompt_text_sql('t.text_value')}=? COLLATE BINARY",
-                (scope,),
+                "pm_prompt_key(t.text_value)=? COLLATE BINARY",
+                (key,),
             )
         ]
     )
 
 
 def _scope_text_id(raw: str | None) -> str | None:
-    """The scope digest of a raw text: that of the text the Library stores."""
-    return None if raw is None else content_digest("text", (stored_prompt_text(raw),))
+    """The scope digest of a raw text: that of its prompt key."""
+    return None if raw is None else content_digest("text", (prompt_key(raw),))
 
 
 def _limit(limit: int) -> None:
@@ -290,7 +290,14 @@ class HistoryReadOperations:
             return None
         text = saved["text"]
         raw_id = _scope_text_id(text)
-        if after is not None and after.raw_text_id != raw_id:
+        # A position from before #460 names the scope by the stripped text
+        # alone; it is the same scope, so a page open across the upgrade resumes.
+        legacy_id = (
+            None
+            if text is None
+            else content_digest("text", (stored_prompt_text(text),))
+        )
+        if after is not None and after.raw_text_id not in (raw_id, legacy_id):
             raise ValueError("History position text scope changed")
         group = {
             "positive_id": positive_id,
@@ -309,7 +316,7 @@ class HistoryReadOperations:
         )
         if raw_id is None:
             return result
-        scope = _scope_text_ids(conn, stored_prompt_text(text))
+        scope = _scope_text_ids(conn, prompt_key(text))
         where = f"r.run_id IN ({_MATCHING})"
         params = [scope]
         if after is not None:
@@ -465,7 +472,7 @@ class HistoryReadOperations:
         if saved["text"] is None:
             return {"run": None}
         epoch = self.get_execution_epoch()
-        stored = stored_prompt_text(saved["text"])
+        key = prompt_key(saved["text"])
         candidates = conn.execute(
             "SELECT r.run_id,rc.anchor_id,rc.context_id,"
             "c.positive_snapshot_id,c.negative_snapshot_id "
@@ -477,7 +484,7 @@ class HistoryReadOperations:
             "AND rc.reference_epoch=? "
             f"ORDER BY {_ORDER},rc.anchor_id COLLATE BINARY LIMIT ?",
             (
-                _scope_text_ids(conn, stored),
+                _scope_text_ids(conn, key),
                 pair_id,
                 epoch,
                 _PAIR_RUN_CANDIDATES,
@@ -489,7 +496,7 @@ class HistoryReadOperations:
             if (
                 positive is None
                 or positive["raw"] is None
-                or stored_prompt_text(positive["raw"]) != stored
+                or prompt_key(positive["raw"]) != key
             ):
                 continue
             hints = _hints(conn, positive["raw"], negative["raw"] if negative else None)

database/library.py

@@ -5,6 +5,7 @@ import sqlite3
 from collections.abc import Iterator
 from contextlib import contextmanager
 
+from .library_folders import folder_condition, list_library_folders
 from .library_tokens import library_edit_token
 
 _ELIGIBLE = """(EXISTS(SELECT 1 FROM prompt_pairs ep WHERE ep.positive_prompt_id=p.id)
@@ -121,7 +122,54 @@ def _archive(value: str) -> str:
     return ARCHIVE_SCOPES[value]
 
 
-def _filter(text: str, category: str | None, tags: list[str]) -> tuple[str, list]:
+# How the Tags filter matches (#444): every tag (AND), any of them (OR), or
+# parents without any tag.
+TAG_MODES = ("all", "any", "untagged")
+# A parent that no tag's Browse returns (#444 "Untagged prompts"): no tag of its
+# own and no active pair carrying one. The vocabulary's untagged count uses it too.
+UNTAGGED = (
+    "NOT EXISTS(SELECT 1 FROM prompt_tags ut WHERE ut.prompt_id=p.id) "
+    "AND NOT EXISTS(SELECT 1 FROM prompt_pairs up JOIN prompt_pair_tags upt "
+    "ON upt.pair_id=up.id WHERE up.positive_prompt_id=p.id AND up.archived=0)"
+)
+
+
+def _tag_mode(tag_mode: object, tags: list[str]) -> str:
+    if not isinstance(tag_mode, str) or tag_mode not in TAG_MODES:
+        raise ValueError("tag_mode must be one of " + ", ".join(TAG_MODES))
+    if tag_mode == "untagged" and tags:
+        raise ValueError("Untagged prompts cannot also filter by tags")
+    return tag_mode
+
+
+def _tag_conditions(tags: list[str], any_tag: bool) -> tuple[list, list, list]:
+    """Parent-arm and pair-arm tag conditions and the tags' parameters."""
+    parent_exists = (
+        "EXISTS(SELECT 1 FROM prompt_tags pt JOIN tags t ON t.id=pt.tag_id "
+        "WHERE pt.prompt_id=p.id AND t.name{test})"
+    )
+    pair_exists = (
+        "EXISTS(SELECT 1 FROM prompt_pair_tags pt JOIN tags t ON t.id=pt.tag_id "
+        "WHERE pt.pair_id=pp.id AND t.name{test})"
+    )
+    if any_tag and tags:
+        # One condition per arm: the card, or one active pair, carries any tag.
+        test = " IN (" + ",".join("?" for _ in tags) + ")"
+        return [parent_exists.format(test=test)], [pair_exists.format(test=test)], tags
+    return (
+        [parent_exists.format(test="=?") for _ in tags],
+        [pair_exists.format(test="=?") for _ in tags],
+        tags,
+    )
+
+
+def _filter(
+    text: str,
+    category: str | None,
+    tags: list[str],
+    tag_mode: str = "all",
+    folder_prefixes: list[str] | None = None,
+) -> tuple[str, list]:
     if not isinstance(text, str) or len(text) > 500:
         raise ValueError("Search text must be a string of at most 500 characters")
     if category is not None and (not isinstance(category, str) or len(category) > 200):
@@ -132,6 +180,8 @@ def _filter(text: str, category: str | None, tags: list[str]) -> tuple[str, list
         or any(not isinstance(tag, str) or not tag or len(tag) > 200 for tag in tags)
     ):
         raise ValueError("Invalid library tags")
+    mode = _tag_mode(tag_mode, tags)
+    untagged = mode == "untagged"
     parent, pair, parent_params, pair_params = [], ["pp.archived=0"], [], []
     if text:
         parent.append("(p.text LIKE ? ESCAPE '\\' OR ls.title LIKE ? ESCAPE '\\')")
@@ -146,17 +196,11 @@ def _filter(text: str, category: str | None, tags: list[str]) -> tuple[str, list
         pair.append("pp.category=?")
         parent_params.append(category)
         pair_params.append(category)
-    for tag in tags:
-        parent.append(
-            "EXISTS(SELECT 1 FROM prompt_tags pt JOIN tags t ON t.id=pt.tag_id "
-            "WHERE pt.prompt_id=p.id AND t.name=?)"
-        )
-        pair.append(
-            "EXISTS(SELECT 1 FROM prompt_pair_tags pt JOIN tags t ON t.id=pt.tag_id "
-            "WHERE pt.pair_id=pp.id AND t.name=?)"
-        )
-        parent_params.append(tag)
-        pair_params.append(tag)
+    parent_tags, pair_tags, tag_params = _tag_conditions(tags, mode == "any")
+    parent.extend(parent_tags)
+    pair.extend(pair_tags)
+    parent_params.extend(tag_params)
+    pair_params.extend(tag_params)
     condition = (
         "("
         + (" AND ".join(parent) or "1=1")
@@ -168,7 +212,17 @@ def _filter(text: str, category: str | None, tags: list[str]) -> tuple[str, list
         + " AND ".join(pair)
         + ")"
     )
-    return f"({_ELIGIBLE}) AND ({condition})", [*parent_params, *pair_params]
+    if untagged:
+        # A whole-card condition: inside the parent-or-pair OR an untagged pair
+        # would let a tagged parent through.
+        condition = f"({condition}) AND {UNTAGGED}"
+    params = [*parent_params, *pair_params]
+    if folder_prefixes:
+        # A whole-card condition around both arms: images link to the parent.
+        folder, folder_params = folder_condition(folder_prefixes)
+        condition = f"({condition}) AND {folder}"
+        params.extend(folder_params)
+    return f"({_ELIGIBLE}) AND ({condition})", params
 
 
 class LibraryOperations:
@@ -251,6 +305,8 @@ class LibraryOperations:
         text: str = "",
         category: str | None = None,
         tags: list[str] | None = None,
+        tag_mode: str = "all",
+        folder_prefixes: list[str] | None = None,
         sort: str = "newest",
         limit: int = 40,
         after: dict | None = None,
@@ -258,7 +314,9 @@ class LibraryOperations:
         _limit(limit)
         if sort not in _ORDERS:
             raise ValueError("Unknown library sort order")
-        where, params = _filter(text, category, [] if tags is None else tags)
+        where, params = _filter(
+            text, category, [] if tags is None else tags, tag_mode, folder_prefixes
+        )
         if after is not None:
             if (
                 not isinstance(after, dict)
@@ -311,6 +369,8 @@ class LibraryOperations:
             "items": rows,
             "total": total,
             "variant_total": variant_total,
+            # Echoed so a client can tell this read applied its tag match.
+            "tag_mode": tag_mode,
             "has_more": has_more,
             "next_key": next_key,
         }
@@ -321,6 +381,8 @@ class LibraryOperations:
         text: str = "",
         category: str | None = None,
         tags: list[str] | None = None,
+        tag_mode: str = "all",
+        folder_prefixes: list[str] | None = None,
         sort: str = "newest",
         limit: int = 40,
         after: dict | None = None,
@@ -333,6 +395,8 @@ class LibraryOperations:
                 text=text,
                 category=category,
                 tags=tags,
+                tag_mode=tag_mode,
+                folder_prefixes=folder_prefixes,
                 sort=sort,
                 limit=limit,
                 after=after,
@@ -463,9 +527,13 @@ class LibraryOperations:
             "categories": [row["category"] for row in rows],
         }
 
-    def get_library_options(self) -> dict:
+    def get_library_options(self, folder_roots=None) -> dict:
+        """Categories; with *folder_roots* (root spellings) also ``folders``."""
         with self.library_snapshot() as (conn, epoch, _ceiling):
-            return self._get_library_options(conn, epoch)
+            result = self._get_library_options(conn, epoch)
+            if folder_roots is not None:
+                result["folders"] = list_library_folders(conn, folder_roots, _ELIGIBLE)
+            return result
 
     def get_library_previews(self, positive_ids: list[int]) -> dict[int, list[dict]]:
         """Read at most two links per parent in one bounded page query."""

database/library_folders.py

@@ -0,0 +1,114 @@
+"""Library output-folder filter (#444): folder names, root prefixes and listing.
+
+A folder is a ``/``-joined path relative to an output root, as Images' folder
+filter and the legacy ``/prompt_manager/subfolders`` name them. A parent
+matches a folder when one of its linked images (``generated_images.prompt_id``,
+the link its previews and image count read) lies in that folder or below it,
+under any spelling of any output root.
+
+Matching compares ``generated_images.file_path``, which is always
+``normcase(normpath(abspath(path)))`` (``models.normalize_image_path``), with
+prefixes built by the same normalization here, so case folding is the
+platform's own on both sides (Windows folds non-ASCII too, unlike SQLite's
+``lower()``) and the comparison is an exact ``substr`` with no wildcards.
+"""
+
+import os
+import sqlite3
+
+_MAX_FOLDER = 1024
+_MAX_PREFIXES = 64
+
+
+def normalize_folder(value: object, pathmod=os.path) -> str:
+    """The folder as ``a/b``, outer slashes dropped; empty means no filter.
+
+    ``/`` always separates. ``\\`` separates only on Windows (``ntpath``): on
+    POSIX it is an ordinary character of a folder's name, and the listing
+    (``relative_folder``) names such a folder with it, so both agree.
+    ``.``/``..``/empty segments, control characters and (on Windows) a drive
+    are refused: the value only ever names a folder below a root.
+    """
+    if not isinstance(value, str) or len(value) > _MAX_FOLDER:
+        raise ValueError(f"folder must be text of at most {_MAX_FOLDER} characters")
+    windows = pathmod.sep == "\\"
+    folder = (value.replace("\\", "/") if windows else value).strip("/")
+    if not folder:
+        return ""
+    parts = folder.split("/")
+    if (
+        any(part in ("", ".", "..") for part in parts)
+        or (windows and ":" in parts[0])
+        or any(ord(char) < 32 or ord(char) == 127 for char in folder)
+    ):
+        raise ValueError("folder must be a folder below an output root")
+    return folder
+
+
+def _base(root: object, pathmod) -> str:
+    return pathmod.normcase(pathmod.normpath(str(root)))
+
+
+def folder_prefixes(roots, folder: str, pathmod=os.path) -> list[str]:
+    """One ``file_path`` prefix per root spelling for a normalized folder."""
+    prefixes = []
+    for root in roots:
+        joined = pathmod.join(_base(root, pathmod), *folder.split("/"))
+        prefix = pathmod.normcase(pathmod.normpath(joined)) + pathmod.sep
+        if prefix not in prefixes:
+            prefixes.append(prefix)
+    if not prefixes or len(prefixes) > _MAX_PREFIXES:
+        raise ValueError("No output folder to filter by")
+    return prefixes
+
+
+def folder_condition(prefixes: list[str]) -> tuple[str, list]:
+    """A condition on alias ``p``: one linked image under any prefix."""
+    tests = " OR ".join("substr(fg.file_path,1,?)=?" for _ in prefixes)
+    params = [value for prefix in prefixes for value in (len(prefix), prefix)]
+    return (
+        "EXISTS(SELECT 1 FROM generated_images fg WHERE fg.prompt_id=p.id "
+        f"AND ({tests}))",
+        params,
+    )
+
+
+def relative_folder(image_path: str, roots, pathmod=os.path) -> str | None:
+    """The folder of *image_path* below the first root spelling holding it.
+
+    The spelling of the stored path is kept (its case on Windows); ``None``
+    for an image directly in a root or outside every root.
+    """
+    directory = pathmod.normpath(pathmod.dirname(image_path))
+    folded = pathmod.normcase(directory)
+    for root in roots:
+        base = _base(root, pathmod)
+        start = base if base.endswith(pathmod.sep) else base + pathmod.sep
+        if not folded.startswith(start):
+            continue
+        # normcase keeps the length for ordinary names; else show the folded form.
+        rest = (directory if len(directory) == len(folded) else folded)[len(start) :]
+        parts = rest.split(pathmod.sep)
+        if any(part in ("", ".", "..") for part in parts):
+            return None
+        return "/".join(parts)
+    return None
+
+
+def list_library_folders(
+    conn: sqlite3.Connection, roots, eligible: str, pathmod=os.path
+) -> list[str]:
+    """Every folder (and ancestor) holding an image of an eligible parent."""
+    folders = set()
+    for (image_path,) in conn.execute(
+        "SELECT DISTINCT g.image_path FROM generated_images g "
+        "JOIN prompts p ON p.id=g.prompt_id WHERE " + eligible
+    ):
+        if not isinstance(image_path, str) or not image_path:
+            continue
+        folder = relative_folder(image_path, roots, pathmod)
+        if folder is None:
+            continue
+        parts = folder.split("/")
+        folders.update("/".join(parts[:index]) for index in range(1, len(parts) + 1))
+    return sorted(folders, key=lambda value: (value.casefold(), value))

database/maintenance.py

@@ -44,6 +44,11 @@ from .health import BATCH_ROWS, link_states, readable_roots, readonly_uri
 from .pairs import PairDependencyError
 from .workflow_documents import documents_of_rows, release_documents
 
+try:
+    from ..utils.hashing import register_prompt_functions
+except ImportError:
+    from utils.hashing import register_prompt_functions
+
 MAINTENANCE_OPERATIONS = (
     "duplicate_prompts",
     "optimize",
@@ -289,6 +294,7 @@ def _read_snapshot(db_path, read):
     conn = sqlite3.connect(readonly_uri(db_path), uri=True)
     try:
         conn.row_factory = sqlite3.Row
+        register_prompt_functions(conn)  # LIVE_HINT_CTES calls them (#460)
         conn.execute("BEGIN")  # one read snapshot for the whole plan
         return read(conn)
     finally:

database/models.py

@@ -38,6 +38,11 @@ from .review_schema import (
     review_without_annotations_error,
 )
 from .saved_views import migrate_saved_views, saved_views_schema_error
+
+try:
+    from ..utils.hashing import register_prompt_functions
+except ImportError:
+    from utils.hashing import register_prompt_functions
 from .image_categories import image_categories_schema_error, migrate_image_categories
 from .workflow_documents import (
     DOCUMENT_ID_COLUMNS,
@@ -153,12 +158,15 @@ def configure_connection(conn: sqlite3.Connection) -> None:
     (common on Windows) let other writers' polling busy handlers starve past
     BUSY_TIMEOUT_SECONDS and fail with "database is locked" (#242). Without
     WAL (for example an in-memory database) SQLite's FULL default is kept.
+    It also registers the prompt-key SQL functions (register_prompt_functions).
     """
     journal_mode = conn.execute("PRAGMA journal_mode = WAL").fetchone()[0]
     conn.execute("PRAGMA foreign_keys = ON")
     conn.execute(f"PRAGMA busy_timeout = {BUSY_TIMEOUT_SECONDS * 1000}")
     if str(journal_mode).lower() == "wal":
         conn.execute("PRAGMA synchronous = NORMAL")
+    # History reads and the live-hint rule call Python's prompt key (#460)
+    register_prompt_functions(conn)
 
 
 class PromptModel:

database/saved_views.py

@@ -33,6 +33,9 @@ MAX_TAG_LENGTH = 50
 _IMAGE_SORTS = ("newest", "oldest", "name", "name-desc", "rating")
 _LIBRARY_SORTS = ("newest", "oldest", "name", "rating")
 _MEDIA = ("all", "image", "video")
+# How a Library view's Tags match (#444): every tag, any of them, or parents
+# without tags ("untagged" never goes with tags; the listing refuses both).
+_TAG_MODES = ("all", "any", "untagged")
 _CONTROL = re.compile(r"[\x00-\x1f\x7f]")
 _SOURCE = re.compile(r"all|0|[1-9][0-9]{0,8}")
 # A collection id as the listing's ``collection`` query takes it (#331): none
@@ -67,6 +70,9 @@ _STATE_FIELDS = {
         "category": ("text", 100, ""),
         "sort": ("choice", _LIBRARY_SORTS, "newest"),
         "tags": ("tags", None, ()),
+        # #444: the Tags match and the output folder (relative, "/"-joined).
+        "tag_mode": ("choice", _TAG_MODES, "all"),
+        "folder": ("text", 1024, ""),
     },
 }
 
@@ -152,6 +158,8 @@ def encode_state(page: object, state: object) -> str:
         _validate_value(key, kind, limit, state[key])
     if state.get("favorites") is True and state.get("collection"):
         raise SavedViewInvalid("A view cannot filter by favorites and a collection")
+    if state.get("tag_mode") == "untagged" and state.get("tags"):
+        raise SavedViewInvalid("A view cannot list untagged prompts and filter tags")
     encoded = json.dumps(
         {key: state[key] for key in fields}, ensure_ascii=False, separators=(",", ":")
     )
@@ -195,6 +203,9 @@ def normalize_state(page: str, raw: object) -> dict:
     # as the listing does for a URL with both (#331 D140).
     if state.get("favorites") is True and state.get("collection"):
         state["favorites"] = False
+    # Never written by this build either: the tags win.
+    if state.get("tag_mode") == "untagged" and state.get("tags"):
+        state["tag_mode"] = "all"
     return state
 
 

database/vocabulary.py

@@ -20,7 +20,7 @@ counted, hashed into the token and renamed with the prompts and pairs.
 
 import sqlite3
 
-from .library import _ELIGIBLE, _limit
+from .library import _ELIGIBLE, UNTAGGED, _escape, _limit
 from .library_commands import _TOKEN, LibraryCommandError, _validate_epoch
 from .library_tokens import canonical_token
 from .models import utc_now_iso
@@ -38,7 +38,7 @@ _IMAGE_FILES = (
 )
 
 
-def _tag_rows(page: str) -> str:
+def _tag_rows(page: str, order: str = "page.id") -> str:
     """Count rows for the registry rows selected by ``page`` (id, name)."""
     # Eligible parent components carrying a tag in the page, plus the
     # positive parent of every active pair carrying it (always eligible).
@@ -57,11 +57,45 @@ def _tag_rows(page: str) -> str:
         "COALESCE(MAX(images.n),0) AS image_count "
         "FROM page LEFT JOIN owners ON owners.tag_id=page.id "
         "LEFT JOIN images ON images.tag_id=page.id "
-        "GROUP BY page.id ORDER BY page.id"
+        "GROUP BY page.id ORDER BY " + order
     )
 
 
 _TAG_PAGE = _tag_rows("SELECT id,name FROM tags WHERE id>? ORDER BY id LIMIT ?")
+# #444 tag sorts: sort -> (key expression on tags/counted, direction). Ties and
+# "created" follow the id, so every order is total and keyset pages are exact.
+TAG_SORTS = {
+    "created": (None, "ASC"),
+    "name": ("name COLLATE NOCASE", "ASC"),
+    "name-desc": ("name COLLATE NOCASE", "DESC"),
+    "prompts": ("prompt_count", "DESC"),
+    "prompts-asc": ("prompt_count", "ASC"),
+}
+_MAX_SEARCH = 200
+_TAG_SEARCH = "name LIKE ? ESCAPE '\\'"
+# Every matching tag ranked by the prompt count its row shows (the owners rule
+# of ``_tag_rows``), then one keyset page of it with its image counts.
+_RANKED_TAG_PAGE = (
+    "WITH owners(tag_id,pid) AS ("
+    "SELECT pt.tag_id,p.id FROM prompts p "
+    "JOIN prompt_tags pt ON pt.prompt_id=p.id WHERE " + _ELIGIBLE + " "
+    "UNION SELECT ppt.tag_id,pp.positive_prompt_id FROM prompt_pairs pp "
+    "JOIN prompt_pair_tags ppt ON ppt.pair_id=pp.id WHERE pp.archived=0),"
+    "counted AS (SELECT t.id,t.name,COUNT(DISTINCT o.pid) AS prompt_count "
+    "FROM tags t LEFT JOIN owners o ON o.tag_id=t.id WHERE {search} "
+    "GROUP BY t.id),"
+    "page AS (SELECT id,name,prompt_count FROM counted WHERE {after} "
+    "ORDER BY prompt_count {direction},id LIMIT ?),"
+    "images(tag_id,n) AS (" + _IMAGE_FILES + " "
+    "WHERE mit.tag_id IN (SELECT id FROM page) GROUP BY mit.tag_id) "
+    "SELECT page.id,page.name,page.prompt_count,"
+    "COALESCE(images.n,0) AS image_count FROM page "
+    "LEFT JOIN images ON images.tag_id=page.id "
+    "ORDER BY page.prompt_count {direction},page.id"
+)
+_UNTAGGED_COUNT = (
+    "SELECT COUNT(*) FROM prompts p WHERE (" + _ELIGIBLE + ") AND " + UNTAGGED
+)
 _TAG_ONE = _tag_rows("SELECT id,name FROM tags WHERE id=?")
 # Every owner kind a tag can sit on, for the delete confirmation.
 _TAG_IMPACT = (
@@ -210,6 +244,28 @@ def _tag_after(after: object) -> int:
     return after
 
 
+def _tag_search(search: object) -> str:
+    if not isinstance(search, str) or len(search) > _MAX_SEARCH:
+        raise ValueError(f"Tag search must be text of at most {_MAX_SEARCH} characters")
+    return search.strip()
+
+
+def _sorted_tag_after(after: object, sort: str) -> tuple | None:
+    """The ``[key, id]`` cursor of a name or prompt-count sort."""
+    if after is None:
+        return None
+    key_type = str if sort.startswith("name") else int
+    if (
+        not isinstance(after, list)
+        or len(after) != 2
+        or type(after[0]) is not key_type
+        or type(after[1]) is not int
+        or after[1] <= 0
+    ):
+        raise ValueError("Invalid vocabulary cursor")
+    return after[0], after[1]
+
+
 def _category_after(after: object) -> tuple[int, str] | None:
     if after is None:
         return None
@@ -431,22 +487,80 @@ class VocabularyOperations:
         epoch: str,
         *,
         limit: int = 100,
-        after: int | None = None,
+        after: object = None,
+        search: str = "",
+        sort: str = "created",
     ) -> dict:
+        """One page of tags matching ``search`` (a literal, case-insensitive
+        substring) in ``sort`` order, with the untagged parents' count."""
         _limit(limit)
-        start = _tag_after(after)
-        total = conn.execute("SELECT COUNT(*) FROM tags").fetchone()[0]
-        rows = [dict(row) for row in conn.execute(_TAG_PAGE, (start, limit + 1))]
+        search = _tag_search(search)
+        if not isinstance(sort, str) or sort not in TAG_SORTS:
+            raise ValueError("sort must be one of " + ", ".join(TAG_SORTS))
+        where, params = "1=1", []
+        if search:
+            where, params = _TAG_SEARCH, [f"%{_escape(search)}%"]
+        total, untagged = conn.execute(
+            f"SELECT (SELECT COUNT(*) FROM tags WHERE {where}),({_UNTAGGED_COUNT})",
+            params,
+        ).fetchone()
+        rows = [
+            dict(row) for row in self._tag_page(conn, where, params, sort, after, limit)
+        ]
         has_more = len(rows) > limit
         rows = rows[:limit]
+        key, _direction = TAG_SORTS[sort]
+        next_key = None
+        if has_more:
+            last = rows[-1]
+            next_key = (
+                last["id"]
+                if key is None
+                else [
+                    last["prompt_count" if key == "prompt_count" else "name"],
+                    last["id"],
+                ]
+            )
         return {
             "library_epoch": epoch,
             "tags": rows,
             "total": total,
+            "untagged_count": untagged,
+            # Echoed so a client can tell this read applied its search and sort.
+            "search": search,
+            "sort": sort,
             "has_more": has_more,
-            "next_key": rows[-1]["id"] if has_more else None,
+            "next_key": next_key,
         }
 
+    @staticmethod
+    def _tag_page(conn, where, params, sort, after, limit):
+        key, direction = TAG_SORTS[sort]
+        if key is None:
+            page = (
+                f"SELECT id,name FROM tags WHERE {where} AND id>? ORDER BY id LIMIT ?"
+            )
+            return conn.execute(
+                _tag_rows(page), [*params, _tag_after(after), limit + 1]
+            )
+        cursor = _sorted_tag_after(after, sort)
+        operator = ">" if direction == "ASC" else "<"
+        keyset, keys = "1=1", []
+        if cursor is not None:
+            keyset = f"({key}{operator}? OR ({key}=? AND id>?))"
+            keys = [cursor[0], cursor[0], cursor[1]]
+        if key == "prompt_count":
+            sql = _RANKED_TAG_PAGE.format(
+                search=where, after=keyset, direction=direction
+            )
+            return conn.execute(sql, [*params, *keys, limit + 1])
+        page = (
+            f"SELECT id,name FROM tags WHERE {where} AND {keyset} "
+            f"ORDER BY {key} {direction},id LIMIT ?"
+        )
+        order = f"page.{key} {direction},page.id"
+        return conn.execute(_tag_rows(page, order), [*params, *keys, limit + 1])
+
     def _vocabulary_categories(
         self,
         conn: sqlite3.Connection,
@@ -490,9 +604,18 @@ class VocabularyOperations:
             ),
         }
 
-    def list_vocabulary_tags(self, *, limit: int = 100, after: int | None = None):
+    def list_vocabulary_tags(
+        self,
+        *,
+        limit: int = 100,
+        after: object = None,
+        search: str = "",
+        sort: str = "created",
+    ):
         with self.library_snapshot() as (conn, epoch, _ceiling):
-            return self._vocabulary_tags(conn, epoch, limit=limit, after=after)
+            return self._vocabulary_tags(
+                conn, epoch, limit=limit, after=after, search=search, sort=sort
+            )
 
     def list_vocabulary_categories(
         self, *, limit: int = 100, after: list | None = None

py/api/__init__.py

@@ -40,6 +40,7 @@ from aiohttp import web
 from PIL import Image
 
 from .prompts import PromptRoutesMixin
+from .revalidation import revalidated_response
 from .pairs import PairRoutesMixin
 from .library import LibraryRoutesMixin
 from .vocabulary import VocabularyRoutesMixin
@@ -653,35 +654,39 @@ class PromptManagerAPI(
 
         @routes.get("/prompt_manager/workspace")
         async def serve_workspace_ui(request: web.Request) -> web.Response:
-            return self._serve_html_page("workspace.html", "Workspace")
+            return self._serve_html_page(request, "workspace.html", "Workspace")
 
         @routes.get("/prompt_manager/web")
         async def serve_web_ui(request):
-            return self._serve_html_page("index.html", "Web UI", cache=False)
+            return self._serve_html_page(request, "index.html", "Web UI", cache=False)
 
         @routes.get("/prompt_manager/gallery.html")
         async def serve_gallery_ui(request):
-            return self._serve_html_page("metadata.html", "Gallery")
+            return self._serve_html_page(request, "metadata.html", "Gallery")
 
         @routes.get("/prompt_manager/admin")
         async def serve_admin_ui(request):
-            return self._serve_html_page("admin.html", "Admin UI")
+            return self._serve_html_page(request, "admin.html", "Admin UI")
 
         @routes.get("/prompt_manager/gallery")
         async def serve_gallery_admin_ui(request):
-            return self._serve_html_page("gallery.html", "Gallery")
+            return self._serve_html_page(request, "gallery.html", "Gallery")
 
         # ── Static file serving ───────────────────────────────────────
 
         @routes.get("/prompt_manager/lib/{filepath:.*}")
         async def serve_lib_static(request):
             """Serve static library files (JS, CSS) from web/lib directory."""
-            return self._serve_static_file("lib", request.match_info.get("filepath"))
+            return self._serve_static_file(
+                request, "lib", request.match_info.get("filepath")
+            )
 
         @routes.get("/prompt_manager/js/{filepath:.*}")
         async def serve_js_static(request):
             """Serve static JavaScript files from web/js directory."""
-            return self._serve_static_file("js", request.match_info.get("filepath"))
+            return self._serve_static_file(
+                request, "js", request.match_info.get("filepath")
+            )
 
         # ── Register domain-specific routes from mixins ───────────────
 
@@ -726,8 +731,11 @@ class PromptManagerAPI(
 
         self.logger.info("All routes registered with decorator pattern")
 
-    def _serve_static_file(self, subdir, filepath):
-        """Serve ``web/<subdir>/<filepath>``; 403 outside it, 404 when missing."""
+    def _serve_static_file(self, request, subdir, filepath):
+        """Serve ``web/<subdir>/<filepath>``; 403 outside it, 404 when missing.
+
+        A served file is revalidated on every reuse (#478, ``revalidation``).
+        """
         root = os.path.join(_get_project_root(), "web", subdir)
         exact = STATIC_EXACT_FILES.get((subdir, filepath))
         # An exact file passes the same containment check, its extension only.
@@ -742,10 +750,13 @@ class PromptManagerAPI(
             return web.Response(text="Not Found", status=404)
         with open(file_path, "rb") as f:
             content = f.read()
-        return web.Response(body=content, content_type=content_type)
+        return revalidated_response(request, content, content_type)
 
-    def _serve_html_page(self, filename, title, cache=True):
-        """Serve ``web/<filename>``; 404 when missing, 500 when unreadable."""
+    def _serve_html_page(self, request, filename, title, cache=True):
+        """Serve ``web/<filename>``; 404 when missing, 500 when unreadable.
+
+        A served page is revalidated on every reuse (#478, ``revalidation``).
+        """
         html_path = os.path.join(_get_project_root(), "web", filename)
         try:
             if cache and html_path in self._html_cache:
@@ -762,7 +773,9 @@ class PromptManagerAPI(
                     content_type="text/html",
                     status=404,
                 )
-            return web.Response(text=content, content_type="text/html", charset="utf-8")
+            return revalidated_response(
+                request, content.encode("utf-8"), "text/html", charset="utf-8"
+            )
         except Exception:
             self.logger.exception(f"Failed to load {title}")
             return web.Response(

py/api/diagnostics_report.py

@@ -213,6 +213,8 @@ def public_config():
             "max_image_age_days": GalleryConfig.MAX_IMAGE_AGE_DAYS,
             "max_concurrent_processing": GalleryConfig.MAX_CONCURRENT_PROCESSING,
             "metadata_extraction_timeout": GalleryConfig.METADATA_EXTRACTION_TIMEOUT,
+            # config.json gallery.monitoring.save_foreign_runs (#247)
+            "save_foreign_runs": GalleryConfig.SAVE_FOREIGN_RUNS,
         },
         "integrations": {
             "lora_manager_enabled": bool(IntegrationConfig.LORA_MANAGER_ENABLED),

py/api/history.py

@@ -27,7 +27,11 @@ from .history_current import (
     resolve_current_candidate,
     snapshot_current_output_scope,
 )
-from .inspection_service import CurrentFileChanged, inspect_current_file
+from .inspection_service import (
+    CurrentFileChanged,
+    image_tags_reader,
+    inspect_current_file,
+)
 from .images import image_category_key
 
 try:
@@ -778,6 +782,14 @@ class HistoryRoutesMixin:
                             ),
                             own,
                         ),
+                        # The file's own image tags, keyed as the Tag dialog
+                        # keys them: the normcased realpath (#457).
+                        image_tags=image_tags_reader(
+                            db,
+                            str(resolved.canonical_target),
+                            self.logger,
+                            close=True,
+                        ),
                         recorded=recorded_output_lookup(
                             db, close=True, reference=receipt["reference"]
                         ),

py/api/images.py

@@ -55,6 +55,7 @@ from .inspection_service import (
     IMAGE_EXTENSIONS as _IMAGE_EXTENSIONS,
     SERVABLE_EXTENSIONS,
     VIDEO_EXTENSIONS,
+    image_tags_reader,
     inspect_current_file,
 )
 
@@ -3962,28 +3963,9 @@ class ImageRoutesMixin:
             )
 
     def _inspection_image_tags(self, root: Path, parts: tuple) -> Callable:
-        """The inspector's ``image_tags`` reader for one rooted file (S1).
-
-        The returned function takes the inspection's own stable stat and
-        answers ``current`` only for a byte record verified at exactly that
-        ``(size, mtime_ns)``; a failed read is ``unavailable``, never an error.
-        """
+        """The inspector's ``image_tags`` reader for one rooted file (S1)."""
         locator = os.path.normcase(os.path.join(os.path.realpath(root), *parts))
-
-        def read(size: int, mtime_ns: int) -> dict:
-            try:
-                found = self.db.current_image_tags({locator: (size, mtime_ns)})
-            except Exception:
-                self.logger.warning(
-                    "Inspection image tags could not be read", exc_info=True
-                )
-                return {"state": "unavailable", "tags": []}
-            tags = found.get(locator)
-            if tags is None:
-                return {"state": "none", "tags": []}
-            return {"state": "current", "tags": list(tags)}
-
-        return read
+        return image_tags_reader(self.db, locator, self.logger)
 
     def _read_own_category(self, sources, path):
         """One file's own category, None, or OWN_CATEGORY_UNREAD (blocking).

py/api/inspection_service.py

@@ -93,6 +93,35 @@ def _in_host_temp(path: Path) -> bool:
         return False
 
 
+def image_tags_reader(
+    db: Any, locator: str, logger: Any, *, close: bool = False
+) -> Callable[[int, int], dict]:
+    """The inspector's ``image_tags`` reader for one file *locator* (S1).
+
+    The returned function takes the inspection's own stable stat and answers
+    ``current`` only for a byte record verified at exactly that
+    ``(size, mtime_ns)``; a failed read is ``unavailable``, never an error.
+    ``close`` closes the calling thread's connection afterwards, for routes
+    that close theirs after reading (History).
+    """
+
+    def read(size: int, mtime_ns: int) -> dict:
+        try:
+            found = db.current_image_tags({locator: (size, mtime_ns)})
+        except Exception:
+            logger.warning("Inspection image tags could not be read", exc_info=True)
+            return {"state": "unavailable", "tags": []}
+        finally:
+            if close:
+                db.model.close()
+        tags = found.get(locator)
+        if tags is None:
+            return {"state": "none", "tags": []}
+        return {"state": "current", "tags": list(tags)}
+
+    return read
+
+
 def inspect_current_file(
     path: Path,
     *,

py/api/library.py

@@ -21,9 +21,11 @@ from .pairs import pair_response
 try:
     from ...database.library import ARCHIVE_SCOPES
     from ...database.library_commands import LibraryCommandError
+    from ...database.library_folders import folder_prefixes, normalize_folder
 except ImportError:
     from database.library import ARCHIVE_SCOPES
     from database.library_commands import LibraryCommandError
+    from database.library_folders import folder_prefixes, normalize_folder
 
 _SECRET = os.urandom(32)
 
@@ -283,36 +285,84 @@ class LibraryRoutesMixin:
         outputs = list(self._get_all_output_dirs())
         return self._media_roots(outputs), len(outputs)
 
+    def _library_root_spellings(self) -> list[str]:
+        """Every spelling of each output root (#444 folder filter): the
+        validated, resolved root and each configured or detected path that
+        resolves to it, so images saved through a symlinked root match."""
+        from ..config import GalleryConfig
+
+        roots = [str(root) for root in self._get_all_output_dirs()]
+        candidates = [
+            *GalleryConfig.MONITORING_DIRECTORIES,
+            self._find_comfyui_output_dir(),
+            *self._comfyui_output_dir_list(),
+        ]
+        spellings = []
+        for root in roots:
+            for value in (root, *candidates):
+                if not value or not isinstance(value, (str, os.PathLike)):
+                    continue
+                value = str(value)
+                same = value == root or os.path.normcase(
+                    os.path.realpath(value)
+                ) == os.path.normcase(root)
+                if same and value not in spellings:
+                    spellings.append(value)
+        return spellings
+
     @pair_response
     async def get_library_options(self, request: web.Request) -> web.Response:
-        result = await self._run_in_executor(self.db.get_library_options)
+        include = request.query.get("include", "")
+        if include not in ("", "folders"):
+            raise ValueError("include must be empty or 'folders'")
+        if include:
+            # Folders are read only when Refine opens (a walk of linked paths).
+            roots = await self._run_in_executor(self._library_root_spellings)
+            result = await self._run_in_executor(self.db.get_library_options, roots)
+        else:
+            result = await self._run_in_executor(self.db.get_library_options)
         return web.json_response({"success": True, **result})
 
     @pair_response
     async def list_library(self, request: web.Request) -> web.Response:
+        folder = normalize_folder(request.query.get("folder", ""))
         options = {
             "text": request.query.get("text", ""),
             "category": request.query.get("category"),
             "tags": sorted(set(request.query.getall("tag", []))),
+            "tag_mode": request.query.get("tag_mode", "all"),
             "sort": request.query.get("sort", "newest"),
         }
         limit = _page_limit(request)
         cursor = request.query.get("cursor")
+        roots = (
+            await self._run_in_executor(self._library_root_spellings) if folder else []
+        )
+        prefixes = folder_prefixes(roots, folder) if folder else None
 
         def read() -> tuple[dict, dict]:
             with self.db.library_snapshot(include_ceiling=True) as snapshot:
                 conn, epoch, ceiling = snapshot
+                # The folder and the roots it was resolved under join the scope.
                 scope = _scope(
-                    {"kind": "parents", "projection": 2, "epoch": epoch, **options}
+                    {
+                        "kind": "parents",
+                        "projection": 2,
+                        "epoch": epoch,
+                        **options,
+                        **({"folder": folder, "roots": roots} if folder else {}),
+                    }
                 )
                 result = self.db._list_library(
                     conn,
                     epoch,
                     ceiling,
                     **options,
+                    folder_prefixes=prefixes,
                     limit=limit,
                     after=_decode(cursor, scope),
                 )
+                result["folder"] = folder
                 records = self.db.get_library_previews(
                     [parent["id"] for parent in result["items"]]
                 )

py/api/revalidation.py

@@ -0,0 +1,47 @@
+"""Revalidated responses for the pages and static files PromptManager serves.
+
+``Cache-Control: no-cache`` lets the browser keep a copy but makes it ask the
+server before every reuse, so a PromptManager update reaches the browser on a
+normal reload (#478). The ETag lets an unchanged file answer 304 without a
+body, which keeps that reload cheap.
+
+The tag hashes the bytes rather than the file's mtime and size: an update can
+keep both (an archive extract, a same-length edit within one clock tick), and
+the handlers read the whole file anyway. It is weak because the gzip
+middleware may send the same content in a different encoding.
+"""
+
+import hashlib
+
+from aiohttp import web
+
+REVALIDATE = "no-cache"
+
+
+def entity_tag(body: bytes) -> str:
+    """Weak ETag for ``body``."""
+    return f'W/"{hashlib.sha256(body).hexdigest()[:32]}"'
+
+
+def etag_matches(if_none_match, etag: str) -> bool:
+    """True when an If-None-Match header names ``etag`` (weak comparison)."""
+    if not if_none_match:
+        return False
+    if if_none_match.strip() == "*":
+        return True
+    opaque = etag.removeprefix("W/")
+    return any(
+        candidate.strip().removeprefix("W/") == opaque
+        for candidate in if_none_match.split(",")
+    )
+
+
+def revalidated_response(request, body: bytes, content_type: str, charset=None):
+    """200 with ``body``, or 304 when the request already holds this version."""
+    etag = entity_tag(body)
+    headers = {"Cache-Control": REVALIDATE, "ETag": etag}
+    if etag_matches(request.headers.get("If-None-Match"), etag):
+        return web.Response(status=304, headers=headers)
+    return web.Response(
+        body=body, content_type=content_type, charset=charset, headers=headers
+    )

py/api/settings_routes.py

@@ -8,7 +8,8 @@ is shared with the LoRA integration's save.
 GET also reports the image sources and the monitor (``sources.py``). A save
 that changes the sources runs as a ``source_change`` registry hold, so it is
 refused while a record-changing job (a scan) runs. ``monitoring_enabled``
-takes effect at the next restart. ``POST /prompt_manager/settings/validate-source``
+takes effect at the next restart; ``save_foreign_runs`` (#247) at once, since
+the monitor reads it for every file. ``POST /prompt_manager/settings/validate-source``
 checks one path and writes nothing.
 """
 
@@ -43,6 +44,7 @@ _SETTINGS_FILE_KEYS = {
     "INFINITE_SCROLL": ("web_ui", "infinite_scroll"),
     "WORKER_THREADS": ("performance", "worker_threads"),
     "MONITORING_ENABLED": ("gallery", "monitoring", "enabled"),
+    "SAVE_FOREIGN_RUNS": ("gallery", "monitoring", "save_foreign_runs"),
     "SAFETY_BACKUP": ("maintenance", "safety_backup"),
     "THUMBNAIL_FORMAT": ("thumbnails", "format"),
     "THUMBNAIL_QUALITY": ("thumbnails", "quality"),
@@ -58,7 +60,7 @@ _SETTINGS_FILE_KEYS = {
 # The thumbnail settings a running Convert outdated job fixes (#317).
 _THUMBNAIL_SETTINGS = ("THUMBNAIL_FORMAT", "THUMBNAIL_QUALITY", "THUMBNAIL_EDGE")
 # Attributes that live on GalleryConfig; the others on PromptManagerConfig.
-_GALLERY_SETTINGS = frozenset({"MONITORING_ENABLED"})
+_GALLERY_SETTINGS = frozenset({"MONITORING_ENABLED", "SAVE_FOREIGN_RUNS"})
 # Attributes that live on AutotagConfig, named without the AUTOTAG_ prefix.
 _AUTOTAG_SETTINGS = frozenset(
     name for name in _SETTINGS_FILE_KEYS if name.startswith("AUTOTAG_")
@@ -264,6 +266,7 @@ class SettingsRoutesMixin:
                     "thumbnail_encoders": encoders,
                     "settings": {
                         "monitoring_enabled": GalleryConfig.MONITORING_ENABLED,
+                        "save_foreign_runs": GalleryConfig.SAVE_FOREIGN_RUNS,
                         "result_timeout": PromptManagerConfig.RESULT_TIMEOUT,
                         "webui_display_mode": PromptManagerConfig.WEBUI_DISPLAY_MODE,
                         "infinite_scroll": PromptManagerConfig.INFINITE_SCROLL,
@@ -626,6 +629,10 @@ class SettingsRoutesMixin:
             if not isinstance(data["monitoring_enabled"], bool):
                 raise ValueError("monitoring_enabled must be true or false")
             updates["MONITORING_ENABLED"] = data["monitoring_enabled"]
+        if "save_foreign_runs" in data:
+            if not isinstance(data["save_foreign_runs"], bool):
+                raise ValueError("save_foreign_runs must be true or false")
+            updates["SAVE_FOREIGN_RUNS"] = data["save_foreign_runs"]
         if "safety_backup" in data:
             if not isinstance(data["safety_backup"], bool):
                 raise ValueError("safety_backup must be true or false")

py/api/vocabulary.py

@@ -61,18 +61,25 @@ class VocabularyRoutesMixin:
         )
 
     async def _vocabulary_page(
-        self, request: web.Request, kind: str, reader_name: str, name: bool
+        self,
+        request: web.Request,
+        kind: str,
+        reader_name: str,
+        name: bool,
+        options: dict | None = None,
     ) -> web.Response:
         limit = _vocabulary_limit(request)
         cursor = request.query.get("cursor")
+        options = options or {}
 
         def read() -> dict:
             # The snapshot stays open while the cursor is decoded and encoded,
-            # so its epoch scope is the epoch the page was read from.
+            # so its epoch scope is the epoch the page was read from. A tag
+            # search or sort joins the scope: its cursor fits no other order.
             with self.db.library_snapshot() as (conn, epoch, _ceiling):
-                scope = _scope({"kind": kind, "epoch": epoch})
+                scope = _scope({"kind": kind, "epoch": epoch, **options})
                 result = getattr(self.db, reader_name)(
-                    conn, epoch, limit=limit, after=_decode(cursor, scope)
+                    conn, epoch, limit=limit, after=_decode(cursor, scope), **options
                 )
                 result["next_cursor"] = _encode(
                     result.pop("next_key"), scope, name=name
@@ -84,8 +91,14 @@ class VocabularyRoutesMixin:
 
     @pair_response
     async def list_vocabulary_tags(self, request: web.Request) -> web.Response:
+        options = {
+            "search": request.query.get("search", "").strip(),
+            "sort": request.query.get("sort", "created"),
+        }
+        # Name keys are text: kept server side so the cursor stays small.
+        name = options["sort"].startswith("name")
         return await self._vocabulary_page(
-            request, "vocabulary-tags", "_vocabulary_tags", False
+            request, "vocabulary-tags", "_vocabulary_tags", name, options
         )
 
     @pair_response

py/config.py

@@ -194,6 +194,8 @@ class GalleryConfig:
 
     Attributes:
         MONITORING_ENABLED (bool): Enable/disable automatic image monitoring
+        SAVE_FOREIGN_RUNS (bool): The monitor stores the certain prompt of an
+            output whose workflow has no PromptManager node (#247)
         MONITORING_DIRECTORIES (List[str]): Directories to monitor for new images
         SUPPORTED_EXTENSIONS (List[str]): Image file extensions to process
         PROCESSING_DELAY (float): Delay in seconds before processing new files
@@ -210,6 +212,8 @@ class GalleryConfig:
 
     # Image monitoring settings
     MONITORING_ENABLED = True
+    # Read by the monitor for every file, so a save applies without a restart
+    SAVE_FOREIGN_RUNS = True
     MONITORING_DIRECTORIES = []  # Auto-detect if empty
     SUPPORTED_EXTENSIONS = [".png", ".jpg", ".jpeg", ".webp", ".gif"]
     PROCESSING_DELAY = 2.0  # Seconds to wait before processing new files
@@ -251,6 +255,7 @@ class GalleryConfig:
         return {
             "monitoring": {
                 "enabled": cls.MONITORING_ENABLED,
+                "save_foreign_runs": cls.SAVE_FOREIGN_RUNS,
                 "directories": cls.MONITORING_DIRECTORIES,
                 "extensions": cls.SUPPORTED_EXTENSIONS,
                 "processing_delay": cls.PROCESSING_DELAY,
@@ -303,6 +308,14 @@ class GalleryConfig:
                 config_logger.warning(
                     "Ignoring gallery monitoring enabled: expected true or false"
                 )
+        if "save_foreign_runs" in monitoring:
+            if isinstance(monitoring["save_foreign_runs"], bool):
+                cls.SAVE_FOREIGN_RUNS = monitoring["save_foreign_runs"]
+            else:
+                config_logger.warning(
+                    "Ignoring gallery monitoring save_foreign_runs: expected true "
+                    "or false"
+                )
         if "directories" in monitoring:
             directories = cls._validated_directories(monitoring["directories"])
             if directories is not None:

tests/browser/full-inspector.spec.mjs

@@ -558,7 +558,8 @@ test.describe('full inspector coarse pointer',()=>{
         }
         await openSelected(page);const viewer=page.locator('dialog#viewer-dialog');
         await expect(viewer.locator('.viewer-canvas img')).toBeVisible();
-        for(const target of await viewer.locator('#viewer-toolbar > :is(button,a),.dialog-header > button').all()) {
+        // #450 D156: the view switch is hidden at 700px and below; a shown one is measured like ×.
+        for(const target of await viewer.locator('#viewer-toolbar > :is(button,a),.dialog-header > button:visible').all()) {
             const box=await target.evaluate(node=>{const {width,height}=node.getBoundingClientRect();return {width,height,name:node.getAttribute('aria-label')};});
             expect(box.width,box.name).toBeGreaterThanOrEqual(80);expect(box.height,box.name).toBeGreaterThanOrEqual(80);
         }
@@ -621,6 +622,11 @@ test('viewer matches the concept shell and toolbar order',async({page,request})=
         const production=await page.evaluate(VIEWER_SHELL);
         // C8 rebuilds the details content; until then the shared file inspector marks the aside it renders into.
         production.paths=production.paths.map(path=>path.replace('aside#viewer-details.file-inspector','aside#viewer-details'));
+        // #450 D156: production's view switch sits right before × (hidden at 700px and below); the concept has none.
+        const modePaths=production.paths.filter(path=>path.includes('button#viewer-mode'));
+        expect(modePaths.map(path=>path.split(' > ').slice(2).join(' > '))).toEqual(['button#viewer-mode.icon-btn','button#viewer-mode.icon-btn > svg.icon','button#viewer-mode.icon-btn > svg.icon > path']);
+        expect(production.paths[production.paths.indexOf(modePaths[2])+1]).toBe('dialog#viewer-dialog > div.dialog-header > button#viewer-close');
+        production.paths=production.paths.filter(path=>!modePaths.includes(path));
         // #249 D34: after Next → production appends its own status lines (position, then the transform and slideshow
         // reasons), and Download stays the original-file link; every other control keeps the concept's text and order.
         const appended=production.toolbar.slice(9);

tests/browser/inspect-concept.spec.mjs

@@ -9,7 +9,7 @@ import { parityPng } from './local-inspect-files.mjs';
 // `#metadata` (docs/concepts/workspace/app.js renderMetadata 740–748, selection 1715–1729, style.css 709–722,
 // 1180–1194, 1543–1545, 1559–1564), at 1600x1100 and 390x844 in both themes, in three states: initial, selected (the
 // same synthetic PNG set into both `#local-file` inputs) and viewer-open. The concept is never edited: the recorded
-// differences D60–D66 (TESTING.md) are applied to its measurement, and, where they change copy, to its page before it
+// differences D60–D66 and D155 (TESTING.md) are applied to its measurement, and, where they change copy, to its page before it
 // is measured, so the browser lays the concept out with production's words. After that the diff must be empty.
 // Paired captures are artifacts only; there are no PNG baselines.
 test.use({ locale: 'en-US', timezoneId: 'UTC', reducedMotion: 'reduce' });
@@ -22,7 +22,8 @@ const COPY = {
         'Read embedded metadata in this browser. The file is not uploaded or added to your library.'],
     chooser: ['Choose an image for a local preview', 'Choose an image or video for a local preview'],
     note: ['This mock previews your file locally. Sample metadata is available with “Open sample metadata”.',
-        'Your file stays in this browser. Its metadata opens with “Open file metadata”.'],
+        // D155 (#448): choosing the file opens its metadata.
+        'Your file stays in this browser. Choosing it opens its metadata.'],
     toast: ['Local preview only. No file uploaded; sample metadata is available above.',
         'Local preview only. No file uploaded; metadata is available above.'],
 };
@@ -212,23 +213,29 @@ async function productionStates(page, theme, testInfo, tag) {
     await settle(page);
     const initial = await page.evaluate(measureInspect, '#workspace-main');
     await captureCanonicalState(page, testInfo, `production-initial-${tag}`);
+    // D155 (#448): choosing the file opens the viewer with its details at once (the concept waits for its heading
+    // action), so the viewer state is measured first; the selected page is measured after Escape, with the focus the
+    // close returned to the chooser removed (the concept's page has no focused control there).
     await page.locator('#workspace-main #local-file').setInputFiles(parityPng(FILE));
-    await expect(page.locator('#workspace-main #local-preview img.local-file-preview')).toBeVisible();
-    await page.locator('#workspace-main #local-preview img').evaluate(image => image.decode().catch(() => {}));
-    await expect(page.locator('#workspace-toast')).toHaveText(COPY.toast[1]);
-    await expect(action).toHaveAttribute('aria-disabled', 'false');
-    // D63: a readable PNG shows no status line.
-    await expect(page.locator('#workspace-main #local-preview p.small')).toHaveCount(0);
-    await settle(page);
-    const selected = await page.evaluate(measureInspect, '#workspace-main');
-    await captureCanonicalState(page, testInfo, `production-selected-${tag}`);
-    await action.click();
     await expect(page.locator('dialog#viewer-dialog[open] #viewer-details .panel-heading')).toBeVisible();
     await expect(page.locator('dialog#viewer-dialog[open] #viewer-details .meta-grid').first()).toBeVisible();
     await page.mouse.move(0, 0);
     await settle(page);
     const viewer = await page.evaluate(measureViewer);
     await captureCanonicalState(page, testInfo, `production-viewer-${tag}`);
+    await page.keyboard.press('Escape');
+    await expect(page.locator('dialog#viewer-dialog')).toHaveCount(0);
+    await page.evaluate(() => document.activeElement?.blur());
+    await expect(page.locator('#workspace-main #local-preview img.local-file-preview')).toBeVisible();
+    await page.locator('#workspace-main #local-preview img').evaluate(image => image.decode().catch(() => {}));
+    // D155: no toast; the opened viewer was the confirmation.
+    await expect(page.locator('#workspace-toast')).toHaveText('');
+    await expect(action).toHaveAttribute('aria-disabled', 'false');
+    // D63: a readable PNG shows no status line.
+    await expect(page.locator('#workspace-main #local-preview p.small')).toHaveCount(0);
+    await settle(page);
+    const selected = await page.evaluate(measureInspect, '#workspace-main');
+    await captureCanonicalState(page, testInfo, `production-selected-${tag}`);
     return { initial, selected, viewer };
 }
 

tests/browser/system-concept.spec.mjs

@@ -475,9 +475,9 @@ function measureAppearance(mainSelector) {
 }
 
 // D97: the rows Appearance adds for the legacy gallery settings, after the concept's own, as production renders them.
+// D156 (#450) moved "Collapse viewer details by default" to the Viewer surface (workspace-viewer-mode.spec.mjs).
 const ADDED_APPEARANCE_ROWS = [
     ['card-details', 'Show image details on cards', 'Show each card’s title and file details under the preview.', true],
-    ['viewer-details', 'Collapse viewer details by default', 'Open the image viewer with its details panel hidden.', false],
 ];
 
 /** The concept's Appearance with the recorded deviations applied, given the server's real settings. */
@@ -488,10 +488,11 @@ function mapAppearanceConcept(reference, settings) {
         control: { ...toggleRow.control, id, checked } }));
     return {
         ...reference,
-        // #317: the server's thumbnail settings follow in their own surface ("Thumbnails", README: retained in Appearance).
-        layout: [reference.layout[0], 'section.surface', ...reference.layout.slice(1)],
-        // D97: two added toggles and the "Reset display defaults" operation close the surface.
-        surface: { ...reference.surface, tree: [...reference.surface.tree, 'div.setting-row', 'div.setting-row', 'div.operation'] },
+        // D156: the Viewer surface, then (#317) the server's thumbnail settings in their own surface ("Thumbnails",
+        // README: retained in Appearance).
+        layout: [reference.layout[0], 'section.surface', 'section.surface', ...reference.layout.slice(1)],
+        // D97: one added toggle (D156 moved the other) and the "Reset display defaults" operation close the surface.
+        surface: { ...reference.surface, tree: [...reference.surface.tree, 'div.setting-row', 'div.operation'] },
         rows: [...reference.rows.map(row => row.control?.id === 'timeout'
             // The node search timeout is the server's real `result_timeout`, not the concept's sample 30.
             ? { ...row, control: { ...row.control, value: String(settings.result_timeout) } } : row), ...added],
@@ -538,7 +539,7 @@ test('appearance comparison detects mutations', async ({ page, request }) => {
     // Unmapped, the concept differs only where D97 and the real timeout are recorded.
     expect(diffInventory(concept, production).map(item => item.path.split('.').slice(0, 2).join('.'))
         .filter((path, index, all) => all.indexOf(path) === index).sort())
-        .toEqual(['layout.1', 'layout.2', 'layout.length', 'operations.0', 'operations.length', 'rows.11', 'rows.12', 'rows.13',
+        .toEqual(['layout.1', 'layout.2', 'layout.3', 'layout.length', 'operations.0', 'operations.length', 'rows.11', 'rows.12',
             'rows.length', 'surface.tree'].sort());
     const measure = () => page.evaluate(measureAppearance, '#workspace-main');
     const paths = async () => diffInventory(expected, await measure()).map(item => item.path);
@@ -617,10 +618,21 @@ const WATCHED = new Set(['watching', 'stopping']);
 const REFRESH_FILES = { title: 'Refresh files', description: 'Read every source again to find files added or removed outside ComfyUI.',
     label: 'Refresh' };
 
+// D158 (#247): the row production adds directly below "Watch for new outputs", in that row's markup.
+const FOREIGN_RUNS = { id: 'foreign-runs', label: 'Save prompts from other workflows',
+    description: 'New outputs of workflows without a PromptManager node add their prompt to the library when it is certain.' };
+
+/** D158: the concept's "Watch for new outputs" row as the inserted "Save prompts from other workflows" row. */
+function foreignRunsRow(monitoringRow, settings) {
+    return { ...monitoringRow, text: [['label', FOREIGN_RUNS.label, FOREIGN_RUNS.id], ['p', FOREIGN_RUNS.description, null]],
+        control: { ...monitoringRow.control, id: FOREIGN_RUNS.id, checked: settings.save_foreign_runs } };
+}
+
 /** The concept's Sources with the recorded deviations applied, given the server's real settings and sources. */
 function mapSourcesConcept(reference, { settings, sources }) {
     const [sample] = reference.sources;
     const scan = sample.actions[0];
+    const rowAt = reference.surface.tree.indexOf('div.setting-row');
     return {
         ...reference,
         // The real sources replace the two samples (without the "sample path" note). D98: each configured one has Remove,
@@ -628,13 +640,15 @@ function mapSourcesConcept(reference, { settings, sources }) {
         sources: sources.effective.map(source => ({ ...sample, title: source.label,
             description: `${source.path} · ${WATCH_TEXT[source.watch]}`,
             actions: [{ ...scan, text: WATCHED.has(source.watch) ? 'Scan changes' : 'Scan folder' }, { ...scan, text: 'Remove' }] })),
-        rows: reference.rows.map(row => {
-            if (row.control.id === 'monitoring') return { ...row, control: { ...row.control, checked: settings.monitoring_enabled } };
+        rows: reference.rows.flatMap(row => {
+            if (row.control.id === 'monitoring') return [{ ...row, control: { ...row.control, checked: settings.monitoring_enabled } },
+                foreignRunsRow(row, settings)];
             // D113: the range ends at the server's core count (the concept's fixed 16) and shows the real worker count.
-            return { ...row, control: { ...row.control, max: String(settings.cpu_count), value: String(settings.worker_threads) } };
+            return [{ ...row, control: { ...row.control, max: String(settings.cpu_count), value: String(settings.worker_threads) } }];
         }),
-        // D98: "Refresh files" closes the surface.
-        surface: { ...reference.surface, tree: [...reference.surface.tree, 'div.operation'] },
+        // D158: the inserted row after the first setting row; D98: "Refresh files" closes the surface.
+        surface: { ...reference.surface, tree: [...reference.surface.tree.slice(0, rowAt + 1), 'div.setting-row',
+            ...reference.surface.tree.slice(rowAt + 1), 'div.operation'] },
         operations: [...reference.operations, { ...reference.operations[0], title: REFRESH_FILES.title,
             description: REFRESH_FILES.description, actions: [{ ...reference.operations[0].actions[0], text: REFRESH_FILES.label }] }],
     };
@@ -688,10 +702,12 @@ test('sources comparison detects mutations', async ({ page, request }) => {
     const expected = mapSourcesConcept(concept, server);
     const production = await productionSources(page, 'dark', { notice: true });
     expect(diffInventory(expected, production)).toEqual([]);
-    // Unmapped, the concept differs only in the real sources, the D98 operation and the D113 range.
+    // Unmapped, the concept differs only in the real sources, the D98 operation, the D158 row (which moves the D113
+    // range to the third row) and the D113 range.
     expect(diffInventory(concept, production).map(item => item.path.split('.').slice(0, 2).join('.'))
         .filter((path, index, all) => all.indexOf(path) === index).sort())
-        .toEqual(['operations.2', 'operations.length', 'rows.1', 'sources.0', 'sources.1', 'surface.tree'].sort());
+        .toEqual(['operations.2', 'operations.length', 'rows.1', 'rows.2', 'rows.length', 'sources.0', 'sources.1',
+            'surface.tree'].sort());
     const measure = () => page.evaluate(measureSources, '#workspace-main');
     const paths = async () => diffInventory(expected, await measure()).map(item => item.path);
     // A removed notice and a removed source row.
@@ -705,6 +721,13 @@ test('sources comparison detects mutations', async ({ page, request }) => {
     await page.evaluate(() => { document.querySelector('#workspace-main label[for="monitoring"]').textContent = 'Watch for outputs'; });
     expect(await paths()).toEqual(['rows.0.text.0.1']);
     await page.evaluate(() => { document.querySelector('#workspace-main label[for="monitoring"]').textContent = 'Watch for new outputs'; });
+    // D158: the inserted row is compared too: a missing one and a changed label are caught.
+    const undoForeign = await mutate(page, { remove: ['#workspace-main .setting-row:has(#foreign-runs)'] });
+    expect(await paths()).toEqual(expect.arrayContaining(['rows.length', 'surface.tree.length']));
+    await undoForeign();
+    await page.evaluate(() => { document.querySelector('#workspace-main label[for="foreign-runs"]').textContent = 'Save other prompts'; });
+    expect(await paths()).toEqual(['rows.1.text.0.1']);
+    await page.evaluate(() => { document.querySelector('#workspace-main label[for="foreign-runs"]').textContent = 'Save prompts from other workflows'; });
     await page.evaluate(() => { document.querySelector('#workspace-main .panel-heading button').textContent = 'Add source'; });
     expect(await paths()).toEqual(['heading.action.text']);
     await page.evaluate(() => { document.querySelector('#workspace-main .panel-heading button').textContent = '+ Add source'; });

tests/browser/vocabulary-canonical-comparison.mjs

@@ -29,6 +29,19 @@ export function pageInventoryOf(main) {
     const rows=[...table.querySelectorAll('tbody tr')];
     const firstCell=rows[0]?.querySelector('td');
     const operations=[...categorySurface.querySelectorAll('.operation')].filter(visible);
+    const operationOf=row=>{const h3=row.querySelector('h3'),p=row.querySelector('p'),rowActions=row.querySelector('.row-actions');return {
+            ...css(row,'display','flexDirection','justifyContent','alignItems','columnGap','paddingTop','paddingBottom','borderTopWidth','borderTopStyle','borderTopColor'),
+            height:px(row.getBoundingClientRect().height),h3:{text:h3.textContent,...font(h3)},p:{text:p.textContent,...font(p)},
+            actions:{...css(rowActions,'flexShrink','flexWrap','columnGap'),buttons:[...rowActions.querySelectorAll('button')].map(buttonOf)}};};
+    // D149 (#444): production adds the Untagged prompts row and the tag search and sort above the table; the concept has
+    // neither. The toolbar is compared on its own (toolbarInventoryOf) with the concept's page toolbar as donor.
+    const untaggedRow=tagSurface.querySelector(':scope > .operation');
+    const d149=untaggedRow?{untagged:operationOf(untaggedRow),
+        // It starts where the concept's table starts: right after the paragraph.
+        offset:px(untaggedRow.getBoundingClientRect().top-tagSurface.getBoundingClientRect().top),
+        // The vertical space the two additions take: the table starts this much lower than after the paragraph alone.
+        // (Both offsets rounded first, so offset + shift is exactly the table's rounded offset.)
+        shift:px(wrap.getBoundingClientRect().top-tagSurface.getBoundingClientRect().top)-px(untaggedRow.getBoundingClientRect().top-tagSurface.getBoundingClientRect().top)}:null;
     return {
         heading:{...css(heading,'display','flexDirection','justifyContent','alignItems','columnGap','rowGap','marginBottom'),
             h1:{text:h1.textContent,tabIndex:h1.tabIndex,...font(h1)},
@@ -54,14 +67,26 @@ export function pageInventoryOf(main) {
             rows:rows.map(row=>{const cells=[...row.querySelectorAll('td')];return {cells:cells.length,height:px(row.getBoundingClientRect().height),tagLines:lines(cells[0]),tag:cells[0].textContent,prompts:cells[1].textContent,
                 images:shownText(cells[2]),imagesHidden:cells[2].querySelector('.ca-visually-hidden')?.textContent??null,
                 actions:[...cells[3].querySelectorAll('.row-actions button')].map(node=>node.textContent.trim())};})},
-        operations:operations.map(row=>{const h3=row.querySelector('h3'),p=row.querySelector('p'),rowActions=row.querySelector('.row-actions');return {
-            ...css(row,'display','flexDirection','justifyContent','alignItems','columnGap','paddingTop','paddingBottom','borderTopWidth','borderTopStyle','borderTopColor'),
-            height:px(row.getBoundingClientRect().height),h3:{text:h3.textContent,...font(h3)},p:{text:p.textContent,...font(p)},
-            actions:{...css(rowActions,'flexShrink','flexWrap','columnGap'),buttons:[...rowActions.querySelectorAll('button')].map(buttonOf)}};}),
+        operations:operations.map(operationOf),
+        d149,
         noPageScroll:document.documentElement.scrollWidth<=innerWidth+1,
     };
 }
 
+/** A `search.toolbar`'s tokens (D149): runs in the page on the concept's page toolbar and on the tag search and sort. */
+export function toolbarInventoryOf(root) {
+    const css=(node,...keys)=>{const style=getComputedStyle(node);return Object.fromEntries(keys.map(key=>[key,style[key]]));};
+    const px=value=>Math.round(value);
+    const field=root.querySelector('.search-field'),input=field.querySelector('input'),select=root.querySelector('select');
+    return {...css(root,'display','columnGap','rowGap','flexWrap','alignItems','marginBottom'),
+        field:{...css(field,'display','alignItems','columnGap','flexGrow','minWidth','paddingLeft','paddingRight','borderTopWidth','borderTopStyle',
+            'borderTopColor','borderTopLeftRadius','backgroundColor'),icon:Boolean(field.querySelector(':scope > svg')),height:px(field.getBoundingClientRect().height)},
+        input:{type:input.type,...css(input,'fontSize','minHeight','backgroundColor','borderTopWidth','color')},
+        select:{...css(select,'fontSize','minHeight','backgroundColor','color','borderTopColor','borderTopLeftRadius','paddingLeft'),
+            height:px(select.getBoundingClientRect().height)}};
+}
+
+
 const countOf=(count,noun)=>`${count} ${noun}${count===1?'':'s'}`;
 
 /**
@@ -81,12 +106,17 @@ export function withImageCounts(operations,{images={},only=[]}={}) {
 
 // Recorded deviations only (plan "Mock inventory" + decisions D1-D15 + ledger rulings). Everything else must be equal.
 // `imageCategories` (#332 D143) is the fixture's own image categories, none by default.
-export function mapVocabularyConcept(reference,imageCategories={}) {
+export function mapVocabularyConcept(reference,imageCategories={},{untaggedText=null,shift=0}={}) {
     const operations=reference.operations.map(row=>({...row,p:{...row.p,
         // Ledger ruling: "1 positive prompts" -> singular, consistent with the shared plural helper (#238/#243).
         text:row.p.text.replace(/^1 positive prompts$/,'1 positive prompt')}}));
-    return {...reference,
-        table:{...reference.table,rows:reference.table.rows.map(row=>({...row,
+    // D149 (#444): the Untagged prompts row is drawn as the concept's first category row with its own text and a Browse
+    // action, and the table moves down by the space it and the tag search and sort take.
+    const first=reference.operations[0];
+    const d149=untaggedText===null?null:{untagged:{...first,h3:{...first.h3,text:'Untagged prompts'},p:{...first.p,text:untaggedText}},
+        offset:reference.table.wrap.offset,shift};
+    return {...reference,d149,
+        table:{...reference.table,wrap:{...reference.table.wrap,offset:reference.table.wrap.offset+shift},rows:reference.table.rows.map(row=>({...row,
             // Data, not a deviation: the concept tags each sample image with its prompt's tags; the vocabulary fixture
             // tags no image, so every Images count is 0 (#249 C2 retired #245 D3).
             images:'0'}))},
@@ -94,9 +124,14 @@ export function mapVocabularyConcept(reference,imageCategories={}) {
     };
 }
 
+/** The D149 mapping inputs: the fixture's Untagged prompts count text (none by default) and production's table shift. */
+export function d149Of(actual,untaggedText='0 positive prompts') {
+    return actual.d149?{untaggedText,shift:actual.d149.shift}:{};
+}
+
 /** The page comparison: concept rows are the concept's sample tags; production appends fixture-only tags (D4). */
 export function compareVocabularyPage(reference,actual,imageCategories={}) {
-    const expected=mapVocabularyConcept(reference,imageCategories);
+    const expected=mapVocabularyConcept(reference,imageCategories,d149Of(actual));
     const shared=expected.table.rows.length;
     return {diff:diffInventory(expected,{...actual,table:{...actual.table,rows:actual.table.rows.slice(0,shared)}}),
         extra:actual.table.rows.slice(shared).map(row=>row.tag)};

tests/browser/workspace-collections.spec.mjs

@@ -363,14 +363,26 @@ test('a list read superseded by another read still opens a collection made since
     const fresh = (await added.json()).collection.id;
     await expect(rows(page)).toHaveCount(2);
     const hold = await holdListReads(page);
-    // A rename schedules the 300 ms count re-read; opening "Fresh" first reads the list once more. The timer's read
-    // supersedes that one while both are held.
+    // A confirmed rename schedules the 300 ms count re-read; opening "Fresh" first reads the list once more. The
+    // timer's read supersedes that one while both are held. The rename's answer is held until "Fresh"'s read is, so
+    // the timer starts only then: with the answer let through at once, a slow host could fire the timer before the
+    // hash change, and the page would then wait on the timer's read alone (#459).
+    let releaseRename;
+    const renameHeld = new Promise(resolve => { releaseRename = resolve; });
+    await page.route(`**${API}/${manifest.collections.Landscapes.id}`, async route => {
+        if (route.request().method() !== 'PATCH') return route.fallback();
+        await renameHeld;
+        return route.fallback();
+    });
     await editButton(page).click();
     await nameField(page).fill('Landscape grids');
     await editDialog(page).getByRole('button', { name: 'Save', exact: true }).click();
-    await expect(toast(page)).toHaveText('Collection renamed.');
     await page.evaluate(hash => { window.location.hash = hash; }, `#/collection/${fresh}`);
-    await expect.poll(() => hold.held.length).toBeGreaterThanOrEqual(2);
+    await expect.poll(() => hold.held.length).toBe(1);
+    releaseRename();
+    // The rename landed (the sidebar row has the new name), so the count read is now due.
+    await expect(row(page, 'Landscape grids')).toHaveCount(1);
+    await expect.poll(() => hold.held.length).toBe(2);
     await expect(main(page)).not.toContainText('Collections could not be loaded.');
     await hold.release();
     await expect(heading(page)).toHaveText('Fresh');

tests/browser/workspace-history.spec.mjs

@@ -1,6 +1,7 @@
 import {startCoverage,saveCoverage} from '../../tools/vendor/browser-coverage.mjs';
 import {test,expect} from './fixtures.mjs';
 import {armUnseenRestores,unseenRestore} from './unseen-restore.mjs';
+import {saveImageTags} from './concept-inventory.mjs';
 import {readFile,writeFile} from 'node:fs/promises';
 import {createHash} from 'node:crypto';
 
@@ -495,23 +496,38 @@ test('the inline current file names its prompts apart from the frozen run fields
     expect(await exact(dialog(page).getByRole('region',{name:'Positive prompt',exact:true}).locator('p.prompt-text'))).toBe(before);
 });
 
-// #449 (D154): History's inline current file draws tags in the shared labelled sections, named with its context so they
-// stay apart from the frozen run's fields. The server sends History no image_tags facet today (D88), so the read is
-// given one here to prove the shared render in History.
-test('the inline current file labels its image tags after its negative prompt',async({page})=>{
+// #449 (D154), #457: History's inline current file draws the file's own image tags in the shared labelled section, named
+// with its context so they stay apart from the frozen run's fields. The tags are saved on the real file through the Tag
+// routes and come from History's own current-file read (its `image_tags` facet), as Images reads them.
+test('the inline current file labels its image tags after its negative prompt',async({page,request})=>{
+    const reads=[];
     await page.route(url=>url.pathname.endsWith('/inspection'),async route=>{const response=await route.fetch();const body=await response.json();
+        reads.push(body.image_tags);
         body.prompts={positive:{status:'resolved',text:'current file positive',source:'graph-inferred'},negative:{status:'resolved',text:'current file negative',source:'graph-inferred'}};
-        body.image_tags={state:'current',tags:['history_tag']};
         await route.fulfill({response,json:body});});
-    await open(page,page.fixture.detail_id);await marker(page,page.fixture.runs.shared).click();await expect(dialog(page).locator('[data-receipt]')).toHaveCount(5);
-    await dialog(page).getByRole('button',{name:'Raw',exact:true}).click();
-    await dialog(page).locator('[data-receipt]').first().getByRole('button',{name:'View current file',exact:true}).click();
-    await expect(page.locator('dialog#viewer-dialog')).toBeVisible();await page.keyboard.press('Escape');await expect(dialog(page)).toBeVisible();
     const current=dialog(page).getByRole('region',{name:'Current reported file',exact:true});
     const tags=current.getByRole('region',{name:'Current file image tags',exact:true});
-    await expect(tags.locator('.role-label')).toHaveText('Image tags · 1');await expect(tags.locator('.tags > .chip')).toHaveText(['history_tag']);
+    const showCurrent=async()=>{
+        await open(page,page.fixture.detail_id);await marker(page,page.fixture.runs.shared).click();await expect(dialog(page).locator('[data-receipt]')).toHaveCount(5);
+        await dialog(page).getByRole('button',{name:'Raw',exact:true}).click();
+        await dialog(page).locator('[data-receipt]').first().getByRole('button',{name:'View current file',exact:true}).click();
+        await expect(page.locator('dialog#viewer-dialog')).toBeVisible();await page.keyboard.press('Escape');await expect(dialog(page)).toBeVisible();
+        await expect(current.getByRole('region',{name:'Current file negative prompt',exact:true})).toBeVisible();
+    };
+    // An untagged file draws no image tags section.
+    await showCurrent();await expect(tags).toHaveCount(0);expect(reads.at(-1)).toEqual({state:'none',tags:[]});
+    await dialog(page).getByRole('button',{name:'Close recorded run',exact:true}).click();
+    expect(await saveImageTags(request,'history/current.png',['history_tag','dusk'])).toEqual(['history_tag','dusk']);
+    await showCurrent();expect(reads.at(-1)).toEqual({state:'current',tags:['history_tag','dusk']});
+    await expect(tags.locator('.role-label')).toHaveText('Image tags · 2');await expect(tags.locator('.tags > .chip')).toHaveText(['history_tag','dusk']);
     const negative=current.getByRole('region',{name:'Current file negative prompt',exact:true});
     expect(await negative.evaluate((node,section)=>({after:Boolean(node.compareDocumentPosition(section)&Node.DOCUMENT_POSITION_FOLLOWING),inside:node.contains(section)}),await tags.elementHandle())).toEqual({after:true,inside:false});
+    // The viewer opened from History shares the inspection: the same section, plain chips (no Tags filter there, D88).
+    await dialog(page).locator('[data-receipt]').first().getByRole('button',{name:'View current file',exact:true}).click();
+    const viewerTags=page.locator('dialog#viewer-dialog #viewer-details section.tag-section.image-tags');
+    await expect(viewerTags.locator('.role-label')).toHaveText('Image tags · 2');
+    await expect(viewerTags.locator('span.chip')).toHaveText(['history_tag','dusk']);await expect(viewerTags.locator('button')).toHaveCount(0);
+    await page.keyboard.press('Escape');await expect(dialog(page)).toBeVisible();
 });
 
 // #332 CUJ 5 (MC8, §5.3): History's current-file inspector reads the file's own category with the Images key rule, so

tests/browser/workspace-inspect.spec.mjs

@@ -8,10 +8,11 @@ import { parityPng, localFixture, tinyWebm, sendablePng, corruptPng, disguised,
 
 const workspace = '/prompt_manager/workspace';
 const TOAST = 'Local preview only. No file uploaded; metadata is available above.';
+const NO_TOAST = '';
 const ACTION = 'Open file metadata';
 const COPY = {
     intro: 'Read embedded metadata in this browser. The file is not uploaded or added to your library.',
-    note: 'Your file stays in this browser. Its metadata opens with “Open file metadata”.',
+    note: 'Your file stays in this browser. Choosing it opens its metadata.',
     chooser: 'Choose an image or video for a local preview',
     multiple: 'Drop one file at a time.',
     directory: 'Folders cannot be inspected. Choose one image or video file.',
@@ -53,13 +54,26 @@ async function choose(page, file) {
     await chooser(page).setInputFiles(file);
 }
 
+/** D155: choosing a file opens the viewer on it with its details; waits for that. */
 async function openViewer(page) {
-    await expect(action(page)).toHaveAttribute('aria-disabled', 'false');
-    await action(page).click();
     await expect(viewer(page)).toBeVisible();
     await expect(rail(page).locator('.panel-heading')).toBeVisible();
 }
 
+/** After the viewer was closed, the heading action (D60) opens it again on the current selection. */
+async function reopenViewer(page) {
+    await expect(action(page)).toHaveAttribute('aria-disabled', 'false');
+    await action(page).click();
+    await openViewer(page);
+}
+
+/** Waits for the viewer a selection opened (D155) and closes it, back on the page. */
+async function closeOpened(page) {
+    await openViewer(page);
+    await page.keyboard.press('Escape');
+    await expect(viewer(page)).toHaveCount(0);
+}
+
 /** Dispatches dragover + drop on `selector` with Files built in the page (and optional string items). */
 function drop(page, selector, { files = [], strings = [] } = {}) {
     return page.evaluate(async ({ selector, files, strings }) => {
@@ -152,7 +166,8 @@ test('chooser opens the shared full inspector without a second read', async ({ p
     const box = await image.evaluate(node => ({ height: node.getBoundingClientRect().height, fit: getComputedStyle(node).objectFit, margin: getComputedStyle(node).marginTop }));
     expect(box.height).toBeLessThanOrEqual(400);
     expect(box).toMatchObject({ fit: 'contain', margin: '20px' });
-    await expect(page.locator('#workspace-toast')).toHaveText(TOAST);
+    // D155: the open viewer is the confirmation, so no toast.
+    await expect(page.locator('#workspace-toast')).toHaveText(NO_TOAST);
     await expect(action(page)).toHaveAttribute('aria-disabled', 'false');
     await expect(action(page)).not.toHaveAttribute('aria-describedby', /.*/);
 
@@ -195,18 +210,120 @@ test('chooser opens the shared full inspector without a second read', async ({ p
 
     await page.keyboard.press('Escape');
     await expect(viewer(page)).toHaveCount(0);
-    await expect(action(page)).toBeFocused();
+    // D155: focus returns to the chooser that opened it.
+    await expect(chooser(page)).toBeFocused();
     // Opening the viewer again shows the same record without reading the File a second time.
     await page.evaluate(() => {
         window.__slices = 0;
         const slice = Blob.prototype.slice;
         Blob.prototype.slice = function (...args) { if (this.name === 'f02.png') window.__slices++; return slice.apply(this, args); };
     });
-    await openViewer(page);
+    await reopenViewer(page);
     expect(await railContent(rail(page))).toEqual(served);
     expect(await page.evaluate(() => window.__slices)).toBe(0);
 });
 
+// #448 (D155): choosing or dropping a file opens the shared viewer with its details showing, in one step. Closing it
+// returns to the page with the file still chosen and focus on the chooser, ready for the next file.
+test('D155: choosing a file opens the viewer with its details in one click', async ({ page }) => {
+    await openInspect(page);
+    await choose(page, parityPng('f02.png'));
+    await expect(viewer(page)).toBeVisible();
+    await expect(viewer(page).locator('#viewer-title')).toHaveText('f02.png');
+    await expect(rail(page)).toBeVisible();
+    await expect(rail(page).locator('.prompt-section.positive p.prompt-text')).toHaveText('a red fox in snow, detailed fur');
+    await expect(viewer(page).getByRole('button', { name: /^(Show|Hide) details$/ })).toHaveCount(0);
+    // The open viewer is the confirmation: no toast says the metadata is "available above".
+    await expect(page.locator('#workspace-toast')).toHaveText('');
+    await page.keyboard.press('Escape');
+    await expect(viewer(page)).toHaveCount(0);
+    await expect(chooser(page)).toBeFocused();
+    await expect(preview(page).locator('img.local-file-preview')).toBeVisible();
+    await expect(action(page)).toHaveAttribute('aria-disabled', 'false');
+    // A dropped second file replaces the first and opens at once.
+    await drop(page, '.file-drop', { files: [sendablePng()] });
+    await expect(viewer(page)).toBeVisible();
+    await expect(viewer(page).locator('#viewer-title')).toHaveText('send-me.png');
+    await page.keyboard.press('Escape');
+    await expect(viewer(page)).toHaveCount(0);
+    await expect(chooser(page)).toBeFocused();
+    // The same file again is a new selection and opens again; the heading action reopens the current one.
+    expect(await chooser(page).evaluate(node => { node.dispatchEvent(new MouseEvent('click')); return node.value; })).toBe('');
+    await choose(page, parityPng('f02.png'));
+    await expect(viewer(page).locator('#viewer-title')).toHaveText('f02.png');
+    await page.keyboard.press('Escape');
+    await action(page).click();
+    await expect(viewer(page).locator('#viewer-title')).toHaveText('f02.png');
+    await expect(rail(page)).toBeVisible();
+    await page.keyboard.press('Escape');
+    await expect(action(page)).toBeFocused();
+});
+
+test('D155: Inspect shows the details even when the viewer collapses them by default (amends D97)', async ({ page }) => {
+    await page.addInitScript(() => localStorage.setItem('gallerySettings', JSON.stringify({ sidebarCollapsedByDefault: true })));
+    await openInspect(page);
+    await choose(page, parityPng('f02.png'));
+    await expect(viewer(page)).toBeVisible();
+    await expect(rail(page)).toBeVisible();
+    const toggle = viewer(page).getByRole('button', { name: 'Hide details', exact: true });
+    await expect(toggle).toHaveAttribute('aria-expanded', 'true');
+    await toggle.click();
+    await expect(rail(page)).toBeHidden();
+    await page.keyboard.press('Escape');
+    await action(page).click();
+    await expect(rail(page)).toBeVisible();
+    await expect(viewer(page).getByRole('button', { name: 'Hide details', exact: true })).toHaveAttribute('aria-expanded', 'true');
+});
+
+test('D155: an unreadable file stays on the page with its alert', async ({ page }) => {
+    await page.addInitScript(() => {
+        const slice = Blob.prototype.slice;
+        Blob.prototype.slice = function (...args) {
+            if (this.name === 'gone.png') return { arrayBuffer: () => Promise.reject(new DOMException('The file is gone', 'NotReadableError')) };
+            return slice.apply(this, args);
+        };
+    });
+    await openInspect(page);
+    await choose(page, parityPng('f02.png', 'gone.png'));
+    await expect(status(page)).toHaveText('Embedded metadata could not be read.');
+    await expect(action(page)).toHaveAttribute('aria-disabled', 'false');
+    await expect(viewer(page)).toHaveCount(0);
+});
+
+test('D155: an AVI keeps its note and toast and does not open the viewer', async ({ page }) => {
+    await openInspect(page);
+    // AVI and WMV metadata is never read and the browser cannot play them, so the page keeps the selection's note.
+    const avi = { name: 'clip.avi', mimeType: 'video/x-msvideo', buffer: Buffer.concat([Buffer.from('RIFF'), Buffer.alloc(4), Buffer.from('AVI LIST'), Buffer.alloc(64)]) };
+    await choose(page, avi);
+    await expect(page.locator('#workspace-toast')).toHaveText(TOAST);
+    await expect(action(page)).toHaveAttribute('aria-disabled', 'false');
+    await expect(viewer(page)).toHaveCount(0);
+    await reopenViewer(page);
+    await expect(viewer(page).locator('#viewer-title')).toHaveText('clip.avi');
+});
+
+test('D155: a selection that lands while another dialog is open leaves the viewer closed', async ({ page }) => {
+    await openInspect(page);
+    await page.evaluate(() => { const other = document.createElement('dialog'); other.id = 'other-dialog'; other.textContent = 'Other'; document.body.append(other); other.showModal(); });
+    await choose(page, parityPng('f02.png'));
+    await expect(action(page)).toHaveAttribute('aria-disabled', 'false');
+    await expect(viewer(page)).toHaveCount(0);
+    await page.evaluate(() => document.querySelector('#other-dialog').close());
+    await reopenViewer(page);
+    await expect(viewer(page).locator('#viewer-title')).toHaveText('f02.png');
+});
+
+test('D155: at 390px the opened details are reachable below the media', async ({ page }) => {
+    await page.setViewportSize({ width: 390, height: 844 });
+    await openInspect(page);
+    await choose(page, parityPng('f02.png'));
+    await expect(viewer(page)).toBeVisible();
+    const facts = rail(page).locator('.meta-grid').first();
+    await facts.scrollIntoViewIfNeeded();
+    await expect(facts).toBeInViewport();
+    expect(await viewer(page).evaluate(node => node.scrollWidth <= node.clientWidth + 1)).toBe(true);
+});
+
 // PR L3 Task 10: local roles are the server inspector's (local-resolver.mjs), so for every parity PNG the rail shows the
 // prompts, their status lines and the facts Images shows for the same bytes in the output folder.
 const PARITY_PNGS = ['f01b', 'f02', 'f03', 'f06a', 'f06b', 'f06c', 'f06d', 'f06e', 'f07a', 'f07b', 'f08', 'f09', 'f10', 'f11', 'f12',
@@ -260,7 +377,9 @@ test('drop accepts exactly one File and never navigates', async ({ page }) => {
     await expect(preview(page)).toBeEmpty();
     expect(await drop(page, '.file-drop', { files: [parityPng('f02.png')] })).toEqual({ over: true, drop: true });
     await expect(preview(page).locator('img.local-file-preview')).toBeVisible();
-    await expect(page.locator('#workspace-toast')).toHaveText(TOAST);
+    // D155: the drop opened the viewer; drops are ignored while it is open (D64), so close it first.
+    await closeOpened(page);
+    await expect(page.locator('#workspace-toast')).toHaveText(NO_TOAST);
     // The chooser names the dropped file, never the one it replaced and never "No file chosen".
     expect(await chooser(page).evaluate(node => [...node.files].map(file => file.name))).toEqual(['f02.png']);
     const first = await preview(page).locator('img').getAttribute('src');
@@ -275,7 +394,7 @@ test('drop accepts exactly one File and never navigates', async ({ page }) => {
         await expect(preview(page).locator('img')).toHaveAttribute('src', first);
     }
     await expect(action(page)).toHaveAttribute('aria-disabled', 'false');
-    await openViewer(page);
+    await reopenViewer(page);
     await expect(viewer(page).locator('#viewer-title')).toHaveText('f02.png');
     await expect(page).toHaveURL(/#\/inspect$/);
 });
@@ -285,6 +404,7 @@ test('cancel and same-file reselect', async ({ page }) => {
     await openInspect(page);
     await choose(page, parityPng('f02.png'));
     await expect(preview(page).locator('img.local-file-preview')).toBeVisible();
+    await closeOpened(page);
     const first = await preview(page).locator('img').getAttribute('src');
     // The chooser still names the chosen file, as in the concept; a cancel (no files) keeps the page as it is.
     expect(await chooser(page).evaluate(node => node.files[0]?.name)).toBe('f02.png');
@@ -308,6 +428,7 @@ test('cancel and same-file reselect', async ({ page }) => {
     const state = await urls(page);
     expect(state.created).toHaveLength(2);
     expect(state.revoked).toEqual([first]);
+    // The new selection opened the viewer again (D155).
     await openViewer(page);
     await expect(viewer(page).locator('.media-position')).toHaveText('1 of 1 loaded: f02.png');
 });
@@ -337,6 +458,7 @@ test('preview does not imply metadata support', async ({ page }) => {
     // Text, markup, SVG and a playlist under media names are unsupported and never get a URL.
     const before = (await urls(page)).created.length;
     for (const file of disguised()) {
+        // Unsupported: nothing opens (D155).
         await choose(page, file);
         await expect(status(page)).toHaveText(COPY.unsupported);
         await expect(status(page)).toHaveAttribute('role', 'alert');
@@ -360,6 +482,7 @@ test('local video plays inline and in the viewer with real tags', async ({ page,
     await expect(rail(page).locator('.meta-grid').first().locator('dd').first()).toHaveText('7');
     await expect(viewer(page).locator('#viewer-stage video')).toHaveCount(1);
     await page.keyboard.press('Escape');
+    await expect(viewer(page)).toHaveCount(0);
     await choose(page, tinyWebm());
     await expect(preview(page).locator('video.local-file-preview')).toBeVisible();
     await openViewer(page);
@@ -386,6 +509,9 @@ test('the toast and a rejected drop never contradict the status line', async ({
     // concept's toast interval.
     await choose(page, undecodableMp4());
     const failed = status(page).filter({ hasText: COPY.previewFailed });
+    // D155: the selection opened the viewer; its media failed there too, and the page's alert waits under it.
+    await expect(failed).toHaveCount(1);
+    await closeOpened(page);
     await expect(failed).toHaveCount(1);
     await drop(page, '.file-drop', { files: [parityPng('f02.png'), parityPng('f03.png')] });
     const rejected = status(page).filter({ hasText: COPY.multiple });
@@ -442,7 +568,8 @@ test('URLs and media end with their owning session', async ({ page }) => {
     let state = await urls(page);
     expect(state.revoked).toEqual([state.created[0]]);
     expect(await page.evaluate(() => window.__loads)).toBeGreaterThanOrEqual(1);
-    // Route change.
+    // Route change, from the page (D155: the selection opened the viewer, which makes the sidebar inert).
+    await closeOpened(page);
     await page.locator('aside.sidebar nav').getByRole('link', { name: 'Overview', exact: true }).click();
     await expect(main(page).getByRole('heading', { level: 1, name: 'Overview', exact: true })).toBeVisible();
     state = await urls(page);

tests/browser/workspace-library-filters.spec.mjs

@@ -0,0 +1,143 @@
+// #444 (D150): the Library's output-folder filter and multi-tag OR in "Refine this view", against the real server, and
+// the Refine dialog against the concept's Filters dialog (app.js:1345-1352).
+import {writeFile} from 'node:fs/promises';
+import {startCoverage,saveCoverage} from '../../tools/vendor/browser-coverage.mjs';
+import {test,expect} from './fixtures.mjs';
+import {captureCanonicalState} from './history-canonical-comparison.mjs';
+import {inventoryOf,diffInventory} from './library-editor-canonical-comparison.mjs';
+
+const route='/prompt_manager/workspace#/library';
+const concept='/__test__/canonical/workspace';
+const cards=page=>page.locator('article[data-prompt-id]');
+const card=(page,text)=>cards(page).filter({has:page.getByText(text,{exact:true})});
+const toolbar=page=>page.locator('#workspace-main search.toolbar');
+const filtersButton=page=>toolbar(page).getByRole('button',{name:/^Filters(?: · \d)?$/});
+const refine=page=>page.getByRole('dialog',{name:'Refine this view',exact:true});
+const folder=page=>refine(page).getByRole('combobox',{name:'Folder',exact:true});
+const tags=page=>refine(page).getByRole('textbox',{name:'Tags',exact:true});
+const match=page=>refine(page).getByRole('combobox',{name:'Tag match',exact:true});
+const apply=page=>refine(page).getByRole('button',{name:'Apply filters',exact:true});
+const caption=page=>page.locator('#workspace-main .section-caption > span').first();
+
+test.beforeEach(async({page,request,browserName})=>{
+    await startCoverage(page,browserName);await expect(await request.post('/__test__/reset',{data:{native_library:true}})).toBeOK();
+    page.libraryErrors=[];page.on('pageerror',error=>page.libraryErrors.push(error.message));
+});
+test.afterEach(async({page,browserName})=>{await saveCoverage(page,browserName);expect(page.libraryErrors).toEqual([]);});
+
+async function open(page) {
+    await page.goto(route);await expect(caption(page)).toHaveText(/^3 positive prompts · /);
+}
+/** Open Refine and wait for the folder list it reads on opening. */
+async function openRefine(page) {
+    const folders=page.waitForResponse(item=>new URL(item.url()).pathname==='/prompt_manager/library/options'&&new URL(item.url()).searchParams.get('include')==='folders');
+    await filtersButton(page).click();await expect(refine(page)).toBeVisible();await folders;
+}
+
+test('Refine offers Folder, Tags and Tag match; a folder narrows the shelf to prompts with images there',async({page})=>{
+    await open(page);await openRefine(page);
+    await expect(refine(page).locator('.field label')).toHaveText(['Folder','Tags','Tag match']);
+    // Both roots hold a "shared" folder; the root's own images name no folder.
+    await expect(folder(page).locator('option')).toHaveText(['All output folders','shared']);
+    await expect(match(page).locator('option')).toHaveText(['All of these tags','Any of these tags','Prompts without tags']);
+    await folder(page).selectOption('shared');await apply(page).click();
+    await expect(caption(page)).toHaveText(/^1 positive prompt · /);
+    await expect(cards(page)).toHaveCount(1);await expect(card(page,'Cat portrait')).toHaveCount(1);
+    await expect(filtersButton(page)).toHaveText('Filters · 1');
+    await openRefine(page);await expect(folder(page)).toHaveValue('shared');
+    await refine(page).getByRole('button',{name:'Clear filters',exact:true}).click();
+    await expect(caption(page)).toHaveText(/^3 positive prompts · /);await expect(filtersButton(page)).toHaveText('Filters');
+});
+
+test('Tag match Any shows every prompt carrying one of the tags; changing only the match applies it',async({page})=>{
+    await open(page);await openRefine(page);
+    await tags(page).fill('cats, dogs');await apply(page).click();
+    await expect(caption(page)).toHaveText(/^0 positive prompts · /);await expect(filtersButton(page)).toHaveText('Filters · 1');
+    await openRefine(page);await match(page).selectOption('any');await apply(page).click();
+    await expect(caption(page)).toHaveText(/^2 positive prompts · /);
+    await expect(cards(page)).toHaveCount(2);await expect(card(page,'Cat portrait')).toHaveCount(1);await expect(card(page,'Dog landscape')).toHaveCount(1);
+    await expect(filtersButton(page)).toHaveText('Filters · 2');
+});
+
+test('a saved view keeps its folder and tag match',async({page})=>{
+    await open(page);await openRefine(page);
+    await folder(page).selectOption('shared');await tags(page).fill('cats, dogs');await match(page).selectOption('any');await apply(page).click();
+    await expect(caption(page)).toHaveText(/^1 positive prompt · /);
+    await toolbar(page).getByRole('button',{name:'Save view',exact:true}).click();
+    const save=page.getByRole('dialog',{name:'Save this view',exact:true});
+    await save.getByRole('textbox',{name:'View name',exact:true}).fill('Shared cats or dogs');
+    const created=page.waitForResponse(item=>new URL(item.url()).pathname==='/prompt_manager/saved-views'&&item.request().method()==='POST');
+    await save.getByRole('button',{name:'Save',exact:true}).click();
+    const body=(await created).request().postDataJSON();
+    expect(body.state).toMatchObject({folder:'shared',tag_mode:'any',tags:['cats','dogs']});
+    await page.goto('about:blank');await open(page);
+    await page.getByRole('navigation',{name:'Saved views',exact:true}).getByRole('link',{name:'Shared cats or dogs',exact:true}).click();
+    await expect(caption(page)).toHaveText(/^1 positive prompt · /);await expect(filtersButton(page)).toHaveText('Filters · 3');
+});
+
+test('a chosen folder missing from the folder list stays listed and chosen in Refine',async({page})=>{
+    await open(page);await openRefine(page);await folder(page).selectOption('shared');await apply(page).click();
+    await expect(caption(page)).toHaveText(/^1 positive prompt · /);
+    // The folder has no image any more as far as the list knows (or the read failed): Apply must not drop it.
+    await page.route('**/prompt_manager/library/options?include=folders',route=>route.fulfill({json:{success:true,library_epoch:'x',categories:[],folders:[]}}));
+    await openRefine(page);
+    await expect(folder(page).locator('option')).toHaveText(['All output folders','shared']);await expect(folder(page)).toHaveValue('shared');
+    await apply(page).click();await expect(caption(page)).toHaveText(/^1 positive prompt · /);await expect(filtersButton(page)).toHaveText('Filters · 1');
+});
+
+// Live B1: the server's flat list sorts full paths as text ("-" and "_" before "/"); Refine shows each folder directly
+// under its own parent and keeps case variants apart.
+test('the Folder list nests each folder under its exact parent',async({page})=>{
+    const folders=['CCC','CCC/mabes9sans-dataset','CCC/mabes9sans-dataset-v1','CCC/mabes9sans-dataset/mabes9sans-dataset_IMG_GEN',
+        'FLUX','FLUX/2025-12-11','flux','flux/images','flux-klein','kiko','kiko-v6','kiko/char'].sort();
+    await page.route('**/prompt_manager/library/options?include=folders',route=>route.fulfill({json:{success:true,library_epoch:'x',categories:[],folders}}));
+    await open(page);await openRefine(page);
+    const I='\u00a0\u00a0';
+    await expect(folder(page).locator('option')).toHaveText(['All output folders','CCC',`${I}mabes9sans-dataset`,`${I}${I}mabes9sans-dataset_IMG_GEN`,
+        `${I}mabes9sans-dataset-v1`,'FLUX',`${I}2025-12-11`,'flux',`${I}images`,'flux-klein','kiko',`${I}char`,'kiko-v6']);
+    expect(await folder(page).locator('option').evaluateAll(nodes=>nodes.map(node=>node.value))).toEqual(['','CCC','CCC/mabes9sans-dataset',
+        'CCC/mabes9sans-dataset/mabes9sans-dataset_IMG_GEN','CCC/mabes9sans-dataset-v1','FLUX','FLUX/2025-12-11','flux','flux/images','flux-klein',
+        'kiko','kiko/char','kiko-v6']);
+});
+
+test('an older server that ignores the folder shows an error with Retry, never the whole shelf',async({page})=>{
+    await open(page);
+    await page.route('**/prompt_manager/library?*',async route=>{
+        const response=await route.fetch();const body=await response.json();delete body.folder;delete body.tag_mode;
+        await route.fulfill({response,json:body});
+    });
+    await openRefine(page);await folder(page).selectOption('shared');await apply(page).click();
+    await expect(page.locator('#workspace-main').getByText('Prompt library could not be loaded. Retry.')).toBeVisible();
+    await expect(cards(page)).toHaveCount(0);
+});
+
+// D150: the Refine dialog is the concept's Filters dialog: its Folder select and Tags input, plus Tag match drawn as the
+// concept's own select; the concept's "Prompt evidence" (no data yet) and its prototype note are not shown.
+function mapFiltersDialog(reference,{folders,matchOptions}) {
+    const [conceptFolder,,conceptTags]=reference.fields;
+    return {...reference,
+        fields:[{...conceptFolder,options:folders},{...conceptTags},{...conceptFolder,label:'Tag match',options:matchOptions}],
+        small:[],
+        // Production's Apply reads "Apply filters" (Images and Library share it, #249).
+        actions:{...reference.actions,buttons:reference.actions.buttons.map(item=>item.text==='Apply tag filter'?{...item,text:'Apply filters'}:item)}};
+}
+for(const [width,height] of [[1600,1100],[390,844]])for(const theme of ['dark','light'])test(`Refine dialog matches the concept Filters dialog ${width} ${theme} (D150)`,async({page},testInfo)=>{
+    await page.setViewportSize({width,height});
+    await page.goto('about:blank');await page.goto(`${concept}#library`);await page.getByRole('combobox',{name:'Theme',exact:true}).selectOption(theme);
+    await page.locator('#main').getByRole('button',{name:'Filters',exact:true}).click();await expect(page.locator('#modal')).toBeVisible();
+    const reference=await page.locator('#modal').evaluate(inventoryOf);
+    await captureCanonicalState(page,testInfo,`concept-filters-${width}-${theme}`,'#modal');
+    await open(page);await page.getByRole('combobox',{name:'Theme',exact:true}).selectOption(theme);await openRefine(page);
+    await expect(folder(page).locator('option')).toHaveCount(2);
+    const actual=await refine(page).evaluate(inventoryOf);
+    await captureCanonicalState(page,testInfo,`production-refine-${width}-${theme}`,'dialog[open]');
+    const expected=mapFiltersDialog(reference,{folders:[['','All output folders'],['shared','shared']],matchOptions:[['all','All of these tags'],['any','Any of these tags'],['untagged','Prompts without tags']]});
+    const path=testInfo.outputPath('refine-inventory.json');await writeFile(path,JSON.stringify({reference,actual},null,2));
+    await testInfo.attach('refine-inventory',{path,contentType:'application/json'});
+    expect(reference.fields.map(item=>item.label)).toEqual(['Folder','Prompt evidence','Tags']);
+    expect(diffInventory(expected,actual)).toEqual([]);
+    await refine(page).getByRole('button',{name:'Close dialog',exact:true}).click();
+    await openRefine(page);await folder(page).selectOption('shared');await tags(page).fill('cats, dogs');await match(page).selectOption('any');await apply(page).click();
+    await expect(caption(page)).toHaveText(/^1 positive prompt · /);await page.mouse.move(0,0);
+    await captureCanonicalState(page,testInfo,`production-library-folder-any-${width}-${theme}`,'#workspace-main');
+});

tests/browser/workspace-library.spec.mjs

@@ -57,6 +57,26 @@ test('library groups reusable negatives under one positive without exposing nega
     for(const item of rows.items) await expect(page.locator(`article[data-prompt-id="${item.id}"] .card-meta > span`).nth(1)).toHaveText(`${item.run_count} run${item.run_count===1?'':'s'}`);
 });
 
+// #464: a starved renderer (headless WebKit under CPU load, a background tab) can hold animation frames for seconds;
+// a loaded page of prompts is shown without waiting for one.
+test('a held animation frame does not hide a loaded page of prompts, the first or a later one',async({page,request})=>{
+    await expect(await request.post('/__test__/reset',{data:{native_library:true,extra_prompts:45}})).toBeOK();
+    const {total}=await(await request.get('/prompt_manager/library?limit=100')).json();
+    await page.addInitScript(()=>{
+        const request=window.requestAnimationFrame.bind(window),cancel=window.cancelAnimationFrame.bind(window);
+        const held=new Map(),real=new Map();let holding=true,next=0;
+        window.requestAnimationFrame=callback=>{const id=--next;if(holding)held.set(id,callback);else real.set(id,request(callback));return id;};
+        window.cancelAnimationFrame=id=>{held.delete(id);if(real.has(id))cancel(real.get(id));real.delete(id);};
+        window.__releaseAnimationFrames=()=>{holding=false;for(const [id,callback] of held)real.set(id,request(callback));held.clear();};
+    });
+    await page.goto(route);
+    const more=page.getByRole('button',{name:'Load more prompts',exact:true});
+    await expect(more).toHaveAttribute('aria-disabled','false');await expect(cards(page)).toHaveCount(40);
+    await more.evaluate(node=>node.click());await expect(page.locator('.library-status')).toHaveText('You’re all caught up.');
+    await expect(cards(page)).toHaveCount(total);
+    await page.evaluate(()=>window.__releaseAnimationFrames());await nextFrame(page);
+    await expect(cards(page)).toHaveCount(total);await expect(page.locator('.library-row-spacer')).toHaveCount(0);
+});
 test('history displays the selected full negative beneath its positive and filters one shared physical gallery',async({page,request})=>{
     const rows=await(await request.get('/prompt_manager/library')).json();const cat=rows.items.find(item=>item.text==='Cat portrait');
     const data=await(await request.get(`/prompt_manager/library/${cat.id}`)).json();
@@ -135,7 +155,8 @@ test('Retry re-reads a failed category list with the shelf',async({page})=>{
     await page.locator('#workspace-main search.toolbar').getByRole('button',{name:'Filters',exact:true}).click();
     await refine(page).getByRole('button',{name:'Apply filters',exact:true}).click();await expect(refine(page)).toHaveCount(0);
     const search=page.getByRole('searchbox',{name:'Search workspace',exact:true});await search.fill(' ');
-    await page.waitForTimeout(400);expect(reads).toEqual([]);await expect(cards(page)).toHaveCount(3);
+    // Only the folder list Refine reads on opening (D150), never the shelf.
+    await page.waitForTimeout(400);expect(reads).toEqual(['/prompt_manager/library/options']);await expect(cards(page)).toHaveCount(3);
 });
 
 test('failed library reads retry while late search responses cannot overwrite newer results',async({page})=>{
@@ -180,7 +201,7 @@ test('library toolbar matches the concept and tags live in Refine this view',asy
     // D44: the active Tags filter shows on the button; the category is the toolbar's own and is not counted.
     await expect(toolbar.getByRole('button')).toHaveText(['Save view','Filters · 1']);
     await toolbar.getByRole('button',{name:'Filters · 1',exact:true}).click();
-    await expect(refine(page).locator('.field label')).toHaveText(['Tags']);await expect(refine(page).getByRole('textbox',{name:'Tags',exact:true})).toHaveValue('studio');
+    await expect(refine(page).locator('.field label')).toHaveText(['Folder','Tags','Tag match']);await expect(refine(page).getByRole('textbox',{name:'Tags',exact:true})).toHaveValue('studio');
     await refine(page).getByRole('button',{name:'Clear filters',exact:true}).click();await expect(refine(page)).toHaveCount(0);
     await expect(toolbar.getByRole('button',{name:'Filters',exact:true})).toBeFocused();
     await expect(page.getByRole('combobox',{name:'Filter by category',exact:true})).toHaveValue('');await expect(cards(page)).toHaveCount(3);

tests/browser/workspace-saved-views.spec.mjs

@@ -208,7 +208,7 @@ test('a view stored under older rules still lists and opens with its usable valu
     await raw({ page: 'library', name: '', state: JSON.stringify({ q: 'dog', sort: 'name-desc' }) });
     expect(await listed(request)).toEqual([
         { id: imagesId, page: 'images', name: 'Old cats', state: { ...IMAGE_DEFAULTS, q: 'cat' } },
-        expect.objectContaining({ page: 'library', name: 'Saved view', state: { q: 'dog', category: '', sort: 'newest', tags: [] } }),
+        expect.objectContaining({ page: 'library', name: 'Saved view', state: { q: 'dog', category: '', sort: 'newest', tags: [], tag_mode: 'all', folder: '' } }),
     ]);
     // A state that is no JSON object is never listed, and its link behaves like a missing view.
     const corruptId = await raw({ page: 'images', name: 'Corrupt', state: '{not json' });
@@ -233,7 +233,7 @@ test('a view stored under older rules still lists and opens with its usable valu
 // D153 (#447): the workspace no longer offers "Highest rated"; the API still stores such a view as it was saved.
 test('a view saved under Highest rated opens on Recently added on Images and the Library', async ({ page, request }) => {
     const images = await create(request, { page: 'images', name: 'Rated cats', state: { ...IMAGE_DEFAULTS, q: 'cat', sort: 'rating' } });
-    const library = await create(request, { page: 'library', name: 'Rated dogs', state: { q: 'dog', category: '', sort: 'rating', tags: [] } });
+    const library = await create(request, { page: 'library', name: 'Rated dogs', state: { q: 'dog', category: '', sort: 'rating', tags: [], tag_mode: 'all', folder: '' } });
     expect((await listed(request)).map(view => view.state.sort)).toEqual(['rating', 'rating']);
     const listings = [];
     page.on('request', item => {
@@ -257,7 +257,7 @@ test('a view saved under Highest rated opens on Recently added on Images and the
 
 test('leaving a saved view returns Images and the Library to their own last state (#249 M2)', async ({ page, request }) => {
     const dogs = await create(request, { page: 'images', name: 'Dogs', state: { ...IMAGE_DEFAULTS, q: 'dogs' } });
-    const cats = await create(request, { page: 'library', name: 'Cats', state: { q: 'cat', category: '', sort: 'newest', tags: [] } });
+    const cats = await create(request, { page: 'library', name: 'Cats', state: { q: 'cat', category: '', sort: 'newest', tags: [], tag_mode: 'all', folder: '' } });
     await page.goto(`${workspace}#/images`);
     await search(page).fill('cat');
     await expect(imageCards(page)).toHaveCount(1);
@@ -300,7 +300,7 @@ test('a toolbar change while sources load keeps the saved Source and Folder (#24
 });
 
 test('a saved view left while its list or rename is pending leaves the next page alone', async ({ page, request }) => {
-    const view = await create(request, { page: 'library', name: 'Cats', state: { q: 'cat', category: '', sort: 'newest', tags: [] } });
+    const view = await create(request, { page: 'library', name: 'Cats', state: { q: 'cat', category: '', sort: 'newest', tags: [], tag_mode: 'all', folder: '' } });
     let releaseList = () => {};
     const holdList = async () => {
         const held = new Promise(resolve => { releaseList = resolve; });
@@ -348,7 +348,7 @@ test('an unknown saved view opens Images, and a failed listing offers Retry', as
     await expect(page).toHaveURL(/#\/images$/);
     await expect(main(page).getByRole('heading', { level: 1 })).toHaveText('Images');
 
-    const view = await create(request, { page: 'library', name: 'Cats', state: { q: 'cat', category: '', sort: 'newest', tags: [] } });
+    const view = await create(request, { page: 'library', name: 'Cats', state: { q: 'cat', category: '', sort: 'newest', tags: [], tag_mode: 'all', folder: '' } });
     let failing = true;
     await page.route(`**${API}`, route => failing && route.request().method() === 'GET'
         ? route.fulfill({ status: 500, contentType: 'application/json', body: '{"success":false,"error":"down"}' })
@@ -539,7 +539,7 @@ test('Edit renames an open view, and deletes it only after confirming inside the
 });
 
 test('a deleted Library view returns to the Library, and failed edits say so', async ({ page, request }) => {
-    const view = await create(request, { page: 'library', name: 'Cats', state: { q: 'cat', category: '', sort: 'newest', tags: [] } });
+    const view = await create(request, { page: 'library', name: 'Cats', state: { q: 'cat', category: '', sort: 'newest', tags: [], tag_mode: 'all', folder: '' } });
     await page.goto(`${workspace}#/saved/${view.id}`);
     await expect(heading(page)).toHaveText('Cats');
     const edit = main(page).locator('.heading-actions > button').first();

tests/browser/workspace-system-sources.spec.mjs

@@ -268,6 +268,71 @@ test('Watch for new outputs saves monitoring_enabled; a restart applies it', asy
     await expect(sourceList(page).locator('.operation p')).toHaveText(['output · watching', 'input · watching']);
 });
 
+test('D158: Save prompts from other workflows saves save_foreign_runs live; off while monitoring is off (#247)', async ({ page, request }, testInfo) => {
+    try {
+        const layout = await openSources(page);
+        const rows = layout.locator('section.surface > .setting-row');
+        await expect(rows.locator(':scope > div > label')).toHaveText(['Watch for new outputs', 'Save prompts from other workflows', 'Worker threads']);
+        const foreign = layout.getByRole('switch', { name: 'Save prompts from other workflows', exact: true });
+        const foreignRow = rows.filter({ has: page.locator('#foreign-runs') });
+        await expect(foreignRow.locator(':scope > div > p')).toHaveText('New outputs of workflows without a PromptManager node add their prompt to the library when it is certain.');
+        await expect(foreign).toBeChecked();
+        await expect(foreign).toBeEnabled();
+        // Keyboard on desktop, a tap on the touch project.
+        if (testInfo.project.name === 'touch') await foreign.tap();
+        else {
+            await foreign.focus();
+            await page.keyboard.press('Space');
+        }
+        await expect(foreign).not.toBeChecked();
+        await expect(statusOf(page)).toHaveText('Unsaved changes.');
+        const saved = page.waitForResponse(response => response.request().method() === 'POST' && response.url().endsWith(SETTINGS_URL));
+        await saveButton(page).click();
+        const response = await saved;
+        expect(response.request().postDataJSON()).toEqual(expect.objectContaining({ save_foreign_runs: false }));
+        expect((await response.json()).restart_required_fields).toEqual([]);
+        // Applied live: the monitor reads it for every file, so no restart is named.
+        await expect(statusOf(page)).toHaveText('Saved to the server configuration.');
+        expect((await serverJson(request, SETTINGS_URL)).settings.save_foreign_runs).toBe(false);
+        // With Watch for new outputs drafted off nothing is watched: the row is disabled and its note says why.
+        const monitoring = layout.getByRole('switch', { name: 'Watch for new outputs', exact: true });
+        await monitoring.uncheck();
+        await expect(foreign).toBeDisabled();
+        const note = foreignRow.locator('p.small');
+        await expect(note).toHaveText('Applies while Watch for new outputs is on.');
+        await expect(foreign).toHaveAttribute('aria-describedby', await note.getAttribute('id'));
+        await monitoring.check();
+        await expect(foreign).toBeEnabled();
+        await expect(note).toHaveCount(0);
+        await expect(foreign).not.toHaveAttribute('aria-describedby', /./);
+        await expect(statusOf(page)).toHaveText('No unsaved changes.');
+    } finally {
+        // The server is shared by the spec's tests: put the default back whatever happened.
+        await request.post(SETTINGS_URL, { data: { save_foreign_runs: true } });
+    }
+});
+
+test('D158: saved off while the monitor still watches until a restart, the foreign row can still be set (D128)', async ({ page, request }) => {
+    await expect(await request.post('/__test__/system/monitor', { data: { watching: ['output', 'input'] } })).toBeOK();
+    try {
+        const layout = await openSources(page);
+        await layout.getByRole('switch', { name: 'Watch for new outputs', exact: true }).uncheck();
+        await saveButton(page).click();
+        await expect(sourceList(page).locator('.operation p').first()).toHaveText('output · watching until restart');
+        const foreign = layout.getByRole('switch', { name: 'Save prompts from other workflows', exact: true });
+        await expect(foreign).toBeEnabled();
+        await expect(layout.locator('.setting-row').filter({ has: page.locator('#foreign-runs') }).locator('p.small'))
+            .toHaveText('Applies while Watch for new outputs is on.');
+        await foreign.uncheck();
+        const saved = page.waitForRequest(item => item.method() === 'POST' && item.url().endsWith(SETTINGS_URL));
+        await saveButton(page).click();
+        expect((await saved).postDataJSON()).toEqual({ save_foreign_runs: false, expected_revision: expect.any(String) });
+        await expect(statusOf(page)).toHaveText('Saved to the server configuration.');
+    } finally {
+        await request.post(SETTINGS_URL, { data: { save_foreign_runs: true } });
+    }
+});
+
 test('worker range max is cpu_count', async ({ page, request }) => {
     const { settings } = await serverJson(request, SETTINGS_URL);
     const layout = await openSources(page);

tests/browser/workspace-system.spec.mjs

@@ -287,7 +287,8 @@ test('Appearance › Thumbnails saves the server thumbnail settings and never co
     await expect(await request.post('/__test__/system/jxl-encoder', { data: { available: false } })).toBeOK();
     await expect(await request.post('/__test__/system/legacy-previews', { data: { count: 1 } })).toBeOK();
     await page.goto(`${workspace}#/system/appearance`);
-    const surface = page.locator('#workspace-main .settings-layout > section.surface').nth(1);
+    // #450 D156: the Viewer surface sits between Browsing & display and Thumbnails.
+    const surface = page.locator('#workspace-main .settings-layout > section.surface').nth(2);
     await expect(surface.locator(':scope > h2')).toHaveText('Thumbnails');
     await expect(surface.locator('.setting-row label')).toHaveText(['Thumbnail format', 'Thumbnail quality', 'Thumbnail size limit']);
     const format = surface.locator('#thumbnail-format');

tests/browser/workspace-viewer-mode.spec.mjs

@@ -0,0 +1,142 @@
+// #450 (TESTING.md D156): every viewer has a full-screen toggle beside ✕. Full screen fills the browser window edge to
+// edge, Centered is the concept's framed dialog; the button names the view it switches to. System › Appearance ›
+// Viewer › "Open viewer in" chooses how the viewer opens; the button changes the current view only.
+import { test, expect } from './fixtures.mjs';
+import { seedConcept, openProduction } from './concept-inventory.mjs';
+import { parityPng } from './local-inspect-files.mjs';
+
+test.use({ locale: 'en-US', timezoneId: 'UTC', reducedMotion: 'reduce' });
+
+const CARDS = '.gallery-body > .image-grid > .media-card';
+const viewer = page => page.locator('dialog#viewer-dialog[open]');
+const modeButton = page => viewer(page).locator('.dialog-header > button#viewer-mode');
+const details = page => viewer(page).locator('#viewer-details');
+
+test.beforeEach(async ({ request }, testInfo) => { testInfo.manifest = await seedConcept(request); });
+
+async function openImages(page, { width = 1600, height = 1100, theme = 'dark' } = {}) {
+    await page.setViewportSize({ width, height });
+    await openProduction(page, 'images', theme);
+    await page.locator(CARDS).first().locator('.image-open').click();
+    await expect(viewer(page).locator('#viewer-details .panel-heading')).toBeVisible();
+}
+
+/** The dialog's box and edges against the window. */
+const frame = page => viewer(page).evaluate(node => {
+    const box = node.getBoundingClientRect(), style = getComputedStyle(node);
+    return { left: Math.round(box.left), top: Math.round(box.top), right: Math.round(innerWidth - box.right),
+        bottom: Math.round(document.documentElement.clientHeight - box.bottom), radius: style.borderTopLeftRadius,
+        border: style.borderTopWidth, fills: Math.abs(box.width - document.documentElement.clientWidth) <= 1 };
+});
+
+test('the button beside ✕ switches between Centered and Full screen and names the view it switches to', async ({ page }) => {
+    await openImages(page);
+    const header = await viewer(page).locator('.dialog-header').evaluate(node => [...node.children].map(child => child.id || child.localName));
+    expect(header).toEqual(['div', 'viewer-mode', 'viewer-close']);
+    await expect(modeButton(page)).toHaveAccessibleName('Full screen view');
+    await expect(modeButton(page)).toHaveAttribute('title', 'Full screen view');
+    await expect(modeButton(page).locator('svg.icon')).toHaveCount(1);
+    const centered = await frame(page);
+    expect(centered.left).toBeGreaterThan(0);
+    expect(centered.fills).toBe(false);
+    // Keyboard: the button is reachable and keeps focus after a switch.
+    await viewer(page).locator('#viewer-close').focus();
+    await page.keyboard.press('Shift+Tab');
+    await expect(modeButton(page)).toBeFocused();
+    await page.keyboard.press('Enter');
+    await expect(viewer(page)).toHaveClass(/\bviewer-fullscreen\b/);
+    await expect(modeButton(page)).toBeFocused();
+    await expect(modeButton(page)).toHaveAccessibleName('Centered view');
+    expect(await frame(page)).toEqual({ left: 0, top: 0, right: 0, bottom: 0, radius: '0px', border: '0px', fills: true });
+    // The details stay toggleable in full screen.
+    const toggle = viewer(page).getByRole('button', { name: 'Hide details', exact: true });
+    await expect(toggle).toHaveAttribute('aria-expanded', 'true');
+    await toggle.click();
+    await expect(details(page)).toBeHidden();
+    await viewer(page).getByRole('button', { name: 'Show details', exact: true }).click();
+    await expect(details(page)).toBeVisible();
+    // Back to Centered: today's dialog, with no toggle while the collapse setting is off.
+    await modeButton(page).click();
+    await expect(viewer(page)).not.toHaveClass(/\bviewer-fullscreen\b/);
+    expect(await frame(page)).toEqual(centered);
+    await expect(viewer(page).getByRole('button', { name: /^(Show|Hide) details$/ })).toHaveCount(0);
+    // Navigation keeps the current view; Esc closes from full screen; the next open is Centered again.
+    await modeButton(page).click();
+    await viewer(page).getByRole('button', { name: 'Next media', exact: true }).click();
+    await expect(viewer(page)).toHaveClass(/\bviewer-fullscreen\b/);
+    await viewer(page).getByRole('button', { name: 'Hide details', exact: true }).focus();
+    await page.keyboard.press('Escape');
+    await expect(viewer(page)).toHaveCount(0);
+    await page.locator(CARDS).first().locator('.image-open').click();
+    await expect(viewer(page)).not.toHaveClass(/\bviewer-fullscreen\b/);
+});
+
+test('switching back to Centered keeps the collapse setting\'s toggle and hidden details (D97)', async ({ page }) => {
+    await page.addInitScript(() => localStorage.setItem('gallerySettings', JSON.stringify({ sidebarCollapsedByDefault: true })));
+    await page.setViewportSize({ width: 1600, height: 1100 });
+    await openProduction(page, 'images', 'dark');
+    await page.locator(CARDS).first().locator('.image-open').click();
+    await expect(viewer(page)).toBeVisible();
+    await expect(details(page)).toBeHidden();
+    await modeButton(page).click();
+    await expect(viewer(page).getByRole('button', { name: 'Show details', exact: true })).toBeVisible();
+    await modeButton(page).click();
+    await expect(details(page)).toBeHidden();
+    await expect(viewer(page).getByRole('button', { name: 'Show details', exact: true })).toHaveAttribute('aria-expanded', 'false');
+});
+
+test('System › Appearance › Viewer chooses how the viewer opens; the button changes only the current view', async ({ page }) => {
+    await page.setViewportSize({ width: 1600, height: 1100 });
+    await openProduction(page, 'system/appearance', 'dark');
+    const surface = page.locator('#workspace-main section.surface').filter({ has: page.locator('h2', { hasText: /^Viewer$/ }) });
+    await expect(surface.locator('.setting-row > div > :is(label, h3)')).toHaveText(['Open viewer in', 'Collapse viewer details by default']);
+    const select = surface.getByRole('combobox', { name: 'Open viewer in', exact: true });
+    await expect(select).toHaveValue('centered');
+    await select.selectOption('fullscreen');
+    await page.getByRole('button', { name: 'Save preferences', exact: true }).click();
+    await expect.poll(() => page.evaluate(() => JSON.parse(localStorage.getItem('gallerySettings') || '{}').viewerOpenMode)).toBe('fullscreen');
+    await page.evaluate(() => { location.hash = '#/images'; });
+    await page.locator(CARDS).first().locator('.image-open').click();
+    await expect(viewer(page)).toHaveClass(/\bviewer-fullscreen\b/);
+    await expect(modeButton(page)).toHaveAccessibleName('Centered view');
+    await modeButton(page).click();
+    await expect(viewer(page)).not.toHaveClass(/\bviewer-fullscreen\b/);
+    expect(await page.evaluate(() => JSON.parse(localStorage.getItem('gallerySettings')).viewerOpenMode)).toBe('fullscreen');
+});
+
+test('Inspect opens full screen with its details when that is the default (D155)', async ({ page }) => {
+    await page.addInitScript(() => localStorage.setItem('gallerySettings', JSON.stringify({ viewerOpenMode: 'fullscreen', sidebarCollapsedByDefault: true })));
+    await page.setViewportSize({ width: 1600, height: 1100 });
+    await openProduction(page, 'inspect', 'dark');
+    await page.locator('#local-file').setInputFiles(parityPng('f02.png'));
+    await expect(viewer(page)).toHaveClass(/\bviewer-fullscreen\b/);
+    await expect(details(page)).toBeVisible();
+    await expect(viewer(page).getByRole('button', { name: 'Hide details', exact: true })).toBeVisible();
+});
+
+test('the Library gallery viewer has the button too', async ({ page }, testInfo) => {
+    await page.setViewportSize({ width: 1600, height: 1100 });
+    await openProduction(page, `library/${testInfo.manifest.prompts[0].positive_id}`, 'dark');
+    await page.locator('.media-card .image-open').first().click();
+    await expect(modeButton(page)).toHaveAccessibleName('Full screen view');
+});
+
+test('at 390px the viewer is already full-bleed and the button hides', async ({ page }) => {
+    await openImages(page, { width: 390, height: 844 });
+    await expect(viewer(page).locator('#viewer-mode')).toBeHidden();
+    await expect(viewer(page).locator('#viewer-close')).toBeVisible();
+});
+
+test('full screen holds at 200% zoom and coarse pointers get a 40px target', async ({ page, browserName }) => {
+    await openImages(page);
+    await modeButton(page).click();
+    await page.evaluate(() => { document.documentElement.style.zoom = '2'; });
+    const zoomed = await frame(page);
+    expect(zoomed).toMatchObject({ left: 0, top: 0, right: 0, bottom: 0 });
+    await page.evaluate(() => { document.documentElement.style.zoom = '1'; });
+    if (await page.evaluate(() => matchMedia('(pointer: coarse)').matches)) {
+        const size = await modeButton(page).evaluate(node => [node.getBoundingClientRect().width, node.getBoundingClientRect().height]);
+        expect(Math.min(...size)).toBeGreaterThanOrEqual(40);
+    }
+    test.info().annotations.push({ type: 'engine', description: browserName });
+});

tests/browser/workspace-vocabulary-search.spec.mjs

@@ -0,0 +1,167 @@
+// #444 (D149): Tags & categories tag search, tag sort and the Untagged prompts row, against the real server.
+import {startCoverage,saveCoverage} from '../../tools/vendor/browser-coverage.mjs';
+import {test,expect} from './fixtures.mjs';
+import {armUnseenRestores} from './unseen-restore.mjs';
+import {captureCanonicalState} from './history-canonical-comparison.mjs';
+
+const route='/prompt_manager/workspace#/tags';
+const main=page=>page.locator('#workspace-main');
+const tagSurface=page=>main(page).locator('.two-col > section.surface').first();
+const table=page=>main(page).getByRole('table',{name:'Reusable tags',exact:true});
+const rows=page=>table(page).locator('tbody tr');
+const names=page=>rows(page).locator('td:first-child');
+const tagRow=(page,name)=>rows(page).filter({has:page.locator('td:first-child').getByText(name,{exact:true})});
+const search=page=>page.getByRole('searchbox',{name:'Search tags',exact:true});
+const sort=page=>page.getByRole('combobox',{name:'Sort tags',exact:true});
+const untagged=page=>tagSurface(page).locator('.operation').filter({has:page.getByRole('heading',{level:3,name:'Untagged prompts',exact:true})});
+const status=page=>tagSurface(page).locator('search.vocabulary-tools [role="status"]');
+const shelf=page=>page.locator('article[data-prompt-id]');
+const filtersButton=(page,name)=>page.locator('#workspace-main search.toolbar').getByRole('button',{name,exact:true});
+const refine=page=>page.getByRole('dialog',{name:'Refine this view',exact:true});
+
+async function seed(request,data={}) {
+    const response=await request.post('/__test__/vocabulary/reset',{data});await expect(response).toBeOK();return response.json();
+}
+let manifest;
+test.beforeEach(async({page,request,browserName})=>{
+    await startCoverage(page,browserName);await armUnseenRestores(page);
+    // Two minimal recipes without tags (the concept recipes all carry tags).
+    manifest=await seed(request,{extra_categories:2});
+    page.vocabularyErrors=[];page.on('pageerror',error=>page.vocabularyErrors.push(error.message));
+});
+test.afterEach(async({page,browserName})=>{await saveCoverage(page,browserName);expect(page.vocabularyErrors).toEqual([]);});
+
+async function open(page) {
+    await page.goto(route);await expect(rows(page)).toHaveCount(manifest.tags.length);
+}
+
+test('the tag vocabulary surface orders its paragraph, Untagged prompts, the search and sort, then the table',async({page})=>{
+    await open(page);
+    const order=await tagSurface(page).evaluate(node=>[...node.children].map(child=>child.matches('.operation')?'operation':
+        child.matches('search.toolbar')?'toolbar':child.matches('.table-wrap')?'table':child.tagName.toLowerCase()).slice(0,5));
+    expect(order).toEqual(['h2','p','operation','toolbar','table']);
+    await expect(untagged(page).locator('p')).toHaveText('2 positive prompts');
+    await expect(untagged(page).getByRole('button')).toHaveText(['Browse']);
+    await expect(search(page)).toHaveAttribute('placeholder','Search tags…');
+    await expect(search(page)).toHaveAttribute('maxlength','200');
+    await expect(sort(page).locator('option')).toHaveText(['Oldest first','Name A–Z','Name Z–A','Most prompts','Fewest prompts']);
+    await expect(sort(page)).toHaveValue('created');
+    await expect(tagSurface(page).locator('label.search-field svg, label.search-field .ca-icon').first()).toBeVisible();
+});
+
+test('search narrows the tags on the server, says when nothing matches and keeps the categories',async({page})=>{
+    await open(page);
+    const categories=await main(page).locator('.two-col > section.surface').nth(1).locator('.operation h3').allTextContents();
+    const reads=[];page.on('request',request=>{if(request.url().includes('/vocabulary/tags?'))reads.push(new URL(request.url()).searchParams);});
+    await search(page).fill('LIGHT');
+    await expect(names(page)).toHaveText(['warm_light','soft_light']);
+    await expect(status(page)).toHaveText('2 tags');
+    expect(reads.at(-1).get('search')).toBe('LIGHT');
+    // Literal: an underscore is not a wildcard.
+    await search(page).fill('a_c');await expect(rows(page)).toHaveCount(0);
+    await expect(tagSurface(page).locator('p.empty')).toHaveText('No tags match this search.');
+    await expect(status(page)).toHaveText('No tags match this search.');
+    await expect(main(page).locator('.two-col > section.surface').nth(1).locator('.operation h3')).toHaveText(categories);
+    await search(page).fill('');await expect(rows(page)).toHaveCount(manifest.tags.length);
+    await expect(status(page)).toHaveText(`${manifest.tags.length} tags`);
+});
+
+test('sort orders every tag by name or by prompts, ties in creation order',async({page})=>{
+    await open(page);
+    const created=await names(page).allTextContents();
+    await sort(page).selectOption('name');
+    const byName=[...created].sort((a,b)=>a.toLowerCase()<b.toLowerCase()?-1:a.toLowerCase()>b.toLowerCase()?1:0);
+    await expect(names(page)).toHaveText(byName);
+    await sort(page).selectOption('name-desc');await expect(names(page)).toHaveText([...byName].reverse());
+    await sort(page).selectOption('prompts');
+    await expect(rows(page).first().locator('td:first-child')).toHaveText('cinematic');
+    const counts=(await rows(page).locator('td:nth-child(2)').allTextContents()).map(Number);
+    expect(counts).toEqual([...counts].sort((a,b)=>b-a));
+    await sort(page).selectOption('prompts-asc');
+    const ascending=(await rows(page).locator('td:nth-child(2)').allTextContents()).map(Number);
+    expect(ascending).toEqual([...ascending].sort((a,b)=>a-b));
+    await sort(page).selectOption('created');await expect(names(page)).toHaveText(created);
+});
+
+test('a rename while searching re-reads the same search and sort',async({page})=>{
+    await open(page);
+    await sort(page).selectOption('name');await search(page).fill('light');
+    await expect(names(page)).toHaveText(['soft_light','warm_light']);
+    await tagRow(page,'warm_light').getByRole('button',{name:'Rename / merge',exact:true}).click();
+    const box=page.locator('dialog.vocabulary-dialog');
+    await box.getByLabel('New name; existing names merge',{exact:true}).fill('amber_light');
+    await box.getByRole('button',{name:'Apply',exact:true}).click();await expect(box).toHaveCount(0);
+    await expect(names(page)).toHaveText(['amber_light','soft_light']);
+    await expect(tagRow(page,'amber_light').getByRole('button',{name:'Rename / merge',exact:true})).toBeFocused();
+    await expect(search(page)).toHaveValue('light');await expect(sort(page)).toHaveValue('name');
+});
+
+test('Untagged prompts Browse opens the Library on exactly the prompts without tags',async({page})=>{
+    await open(page);
+    await untagged(page).getByRole('button',{name:'Browse',exact:true}).click();
+    await expect(page).toHaveURL(/#\/library$/);
+    await expect(main(page).getByRole('heading',{level:1})).toBeFocused();
+    await expect(shelf(page)).toHaveCount(2);
+    await expect(page.locator('#workspace-main .section-caption > span').first()).toHaveText(/^2 positive prompts · /);
+    for(const card of await shelf(page).all())await expect(card.locator('.tags .chip')).toHaveCount(0);
+    await filtersButton(page,'Filters · 1').click();
+    await expect(refine(page).getByRole('combobox',{name:'Tag match',exact:true})).toHaveValue('untagged');
+    await expect(refine(page).getByRole('textbox',{name:'Tags',exact:true})).toBeDisabled();
+    // Changing only the Tag match applies it: the whole shelf returns.
+    await refine(page).getByRole('combobox',{name:'Tag match',exact:true}).selectOption('all');
+    await expect(refine(page).getByRole('textbox',{name:'Tags',exact:true})).toBeEnabled();
+    await refine(page).getByRole('button',{name:'Apply filters',exact:true}).click();
+    await expect(filtersButton(page,'Filters')).toBeVisible();
+    await expect.poll(()=>shelf(page).count()).toBeGreaterThan(2);
+});
+
+test('Refine › Prompts without tags on the shelf drops typed tags and Clear restores All',async({page})=>{
+    await page.goto('/prompt_manager/workspace#/library');await expect(shelf(page).first()).toBeVisible();
+    await filtersButton(page,'Filters').click();
+    await refine(page).getByRole('textbox',{name:'Tags',exact:true}).fill('cinematic');
+    await refine(page).getByRole('combobox',{name:'Tag match',exact:true}).selectOption('untagged');
+    await refine(page).getByRole('button',{name:'Apply filters',exact:true}).click();
+    await expect(shelf(page)).toHaveCount(2);
+    await filtersButton(page,'Filters · 1').click();
+    await expect(refine(page).getByRole('textbox',{name:'Tags',exact:true})).toHaveValue('');
+    await refine(page).getByRole('button',{name:'Clear filters',exact:true}).click();
+    await expect(filtersButton(page,'Filters')).toBeVisible();await expect.poll(()=>shelf(page).count()).toBeGreaterThan(2);
+});
+
+test('an older server that ignores the search shows an error with Retry, never an unfiltered list',async({page})=>{
+    await open(page);
+    // An un-restarted server answers without the echo.
+    await page.route('**/prompt_manager/vocabulary/tags?*',async route=>{
+        const response=await route.fetch();const body=await response.json();delete body.search;delete body.sort;
+        await route.fulfill({response,json:body});
+    });
+    await search(page).fill('light');
+    await expect(tagSurface(page).locator('.vocabulary-error')).toBeVisible();
+    await expect(rows(page)).toHaveCount(0);
+    await expect(tagSurface(page).getByRole('button',{name:'Retry',exact:true})).toBeVisible();
+});
+
+test('the search and sort fit 390px without a page scroll',async({page})=>{
+    await page.setViewportSize({width:390,height:844});await open(page);
+    const layout=await tagSurface(page).evaluate(node=>{
+        const tools=node.querySelector('search.vocabulary-tools'),field=tools.querySelector('.search-field'),select=tools.querySelector('select');
+        const box=node.getBoundingClientRect(),fieldBox=field.getBoundingClientRect(),selectBox=select.getBoundingClientRect();
+        return {inside:fieldBox.left>=box.left&&selectBox.right<=box.right+0.5,fullField:Math.round(fieldBox.width)===Math.round(tools.getBoundingClientRect().width)};
+    });
+    expect(layout).toEqual({inside:true,fullField:true});
+    expect(await page.evaluate(()=>document.documentElement.scrollWidth-document.documentElement.clientWidth)).toBeLessThanOrEqual(1);
+});
+
+// Artifact-only captures for the maintainer's quick check (no pixel assertions): a search with a sort, the Library on
+// Untagged prompts and its Refine dialog.
+for(const [width,height] of [[1600,1100],[390,844]])for(const theme of ['dark','light'])test(`D149 captures ${width} ${theme}`,async({page},testInfo)=>{
+    await page.setViewportSize({width,height});await open(page);
+    await page.getByRole('combobox',{name:'Theme',exact:true}).selectOption(theme);
+    await sort(page).selectOption('prompts');await search(page).fill('i');await expect(status(page)).toHaveText(/^\d+ tags$/);
+    await page.mouse.move(0,0);
+    await captureCanonicalState(page,testInfo,`production-tags-search-${width}-${theme}`,'#workspace-main');
+    await untagged(page).getByRole('button',{name:'Browse',exact:true}).click();await expect(shelf(page)).toHaveCount(2);
+    await captureCanonicalState(page,testInfo,`production-library-untagged-${width}-${theme}`,'#workspace-main');
+    await filtersButton(page,'Filters · 1').click();await expect(refine(page)).toBeVisible();
+    await captureCanonicalState(page,testInfo,`production-refine-untagged-${width}-${theme}`,'dialog[open]');
+});

tests/browser/workspace-vocabulary.spec.mjs

@@ -3,7 +3,7 @@ import {writeFile} from 'node:fs/promises';
 import {test,expect} from './fixtures.mjs';
 import {armUnseenRestores,unseenRestore,followRestore,SEEN_RESTORE_PILL,SEEN_RESTORE_TOAST} from './unseen-restore.mjs';
 import {captureCanonicalState} from './history-canonical-comparison.mjs';
-import {pageInventoryOf,inventoryOf,diffInventory,mapVocabularyConcept,mapVocabularyDialog} from './vocabulary-canonical-comparison.mjs';
+import {pageInventoryOf,inventoryOf,diffInventory,mapVocabularyConcept,mapVocabularyDialog,d149Of,toolbarInventoryOf} from './vocabulary-canonical-comparison.mjs';
 
 const route='/prompt_manager/workspace#/tags';
 // Prompts per concept tag (app.js:71-165): first appearance order, Library Browse counts.
@@ -786,6 +786,15 @@ async function openProductionTags(page,theme,motion) {
     await expect(rows(page)).toHaveCount(manifest.tags.length);await expect(categoryRow(page,'Landscapes')).toHaveCount(1);
 }
 const pageOf=(page,production)=>page.locator(production?'#workspace-main':'#main').evaluate(pageInventoryOf);
+/** D149 donor: the concept's own page toolbar (Prompt library), whose search field and select the tag search reuses. */
+async function conceptToolbar(page,theme) {
+    await page.goto('about:blank');await page.goto(`${concept}#library`);
+    await page.getByRole('combobox',{name:'Theme',exact:true}).selectOption(theme);
+    const inventory=await page.locator('#main search.toolbar').evaluate(toolbarInventoryOf);
+    await page.goto('about:blank');await page.goto(`${concept}#tags`);await page.getByRole('combobox',{name:'Theme',exact:true}).selectOption(theme);
+    return inventory;
+}
+const tagToolbar=page=>page.locator('#workspace-main .two-col > section.surface:first-child > search.toolbar').evaluate(toolbarInventoryOf);
 /** Breadcrumb, document title (D7: "Tags", only the concept suffix differs) and the keyboard focus ring on a row action. */
 async function chromeOf(page,production) {
     const scope=page.locator(production?'#workspace-main':'#main');
@@ -805,7 +814,7 @@ async function chromeOf(page,production) {
 }
 /** Concept rows are the concept's own sample tags; production appends the fixture-only tags after them (D4 creation order). */
 function comparePage(reference,actual) {
-    const expected=mapVocabularyConcept(reference);
+    const expected=mapVocabularyConcept(reference,{},d149Of(actual));
     const shared=expected.table.rows.length;
     return {diff:diffInventory(expected,{...actual,table:{...actual.table,rows:actual.table.rows.slice(0,shared)}}),
         extra:actual.table.rows.slice(shared).map(row=>row.tag)};
@@ -818,13 +827,18 @@ for(const [width,height] of SIZES)for(const theme of ['dark','light'])test(`tags
         await openConceptTags(page,theme,motion);const reference=await pageOf(page,false);
         if(motion==='no-preference')await captureCanonicalState(page,testInfo,`concept-tags-${width}-${theme}`,'#main');
         const referenceChrome=await chromeOf(page,false);
+        const referenceTools=await conceptToolbar(page,theme);
         await openProductionTags(page,theme,motion);const actual=await pageOf(page,true);
+        const actualTools=await tagToolbar(page);
         if(motion==='no-preference')await captureCanonicalState(page,testInfo,`production-tags-${width}-${theme}`,'#workspace-main');
-        result[motion]={reference,actual,...comparePage(reference,actual),referenceChrome,actualChrome:await chromeOf(page,true)};
+        result[motion]={reference,actual,...comparePage(reference,actual),referenceChrome,actualChrome:await chromeOf(page,true),
+            toolbarDiff:diffInventory(referenceTools,actualTools)};
     }
     const path=testInfo.outputPath('tags-geometry.json');await writeFile(path,JSON.stringify(result,null,2));
     await testInfo.attach('tags-geometry',{path,contentType:'application/json'});
-    for(const [motion,{reference,diff,extra,referenceChrome,actualChrome}] of Object.entries(result)) {
+    for(const [motion,{reference,diff,extra,referenceChrome,actualChrome,toolbarDiff}] of Object.entries(result)) {
+        // D149 (#444): the tag search and sort carry the concept page toolbar's tokens.
+        expect.soft(toolbarDiff,`${motion} tag search and sort against the concept toolbar`).toEqual([]);
         expect.soft(referenceChrome,`${motion} concept chrome`).toMatchObject({breadcrumb:'Tags',title:'Tags',ring:{visible:true},linkRing:{outlineOffset:'3px'}});
         expect.soft(actualChrome,`${motion} breadcrumb, title and focus ring`).toEqual(referenceChrome);
         expect.soft(reference.heading.actions.buttons.map(item=>item.text),`${motion} concept heading actions`).toEqual(['+ Create tag','Auto-tag images']);
@@ -894,6 +908,13 @@ test('mutation check: the comparison reports a changed surface padding and a cha
     expect(paths).toContain('surfaces.1.paddingLeft');
     expect(paths).toContain('table.rows.0.actions.0');
     expect(paths).toContain('table.rowActions.buttons.0.text');
+    // D149: a moved Untagged prompts row and a restyled tag search are reported too.
+    await page.addStyleTag({content:'.vocabulary-main .surface > .operation{margin-top:6px!important}.vocabulary-tools .search-field{border-radius:9px!important}'});
+    const moved=comparePage(reference,await pageOf(page,true)).diff.map(item=>item.path);
+    expect(moved).toContain('d149.offset');
+    const tools=await conceptToolbar(page,'dark');await openProductionTags(page,'dark','no-preference');
+    await page.addStyleTag({content:'.vocabulary-tools .search-field{border-radius:9px!important}'});
+    expect(diffInventory(tools,await tagToolbar(page)).map(item=>item.path)).toContain('field.borderTopLeftRadius');
     // The dialog comparison is not vacuous either: a removed Cancel and a recolored field are reported.
     const donors=await conceptDonors(page,'dark');
     await openConceptTags(page,'dark','no-preference');await conceptDialogs.create(page);

tests/browser/workspace.spec.mjs

@@ -472,7 +472,7 @@ test('System is native: the concept shell, section navigation, sections and the
         if (hash === 'overview') await expect(layout.locator(':scope > .summary-metrics')).toBeVisible();
         if (hash === 'sources') await expect(layout.locator('#source-list .operation')).toHaveCount(2);
         expect(await layout.evaluate(node => [...node.children].map(child => `${child.localName}.${child.className}`)))
-            .toEqual({ overview: ['div.summary-metrics', 'div.two-col', 'div.save-row'], appearance: ['section.surface', 'section.surface', 'div.save-row'],
+            .toEqual({ overview: ['div.summary-metrics', 'div.two-col', 'div.save-row'], appearance: ['section.surface', 'section.surface', 'section.surface', 'div.save-row'] /* D156: Browsing & display, Viewer, Thumbnails */,
                 sources: ['div.notice', 'section.surface', 'div.save-row'],
                 integrations: ['section.surface', 'section.surface', 'div.save-row'],
                 maintenance: ['div.notice', 'section.surface', 'section.surface', 'div.save-row'],

tests/comfyui/execution-history.spec.mjs

@@ -314,6 +314,16 @@ test('fresh registered PM variants preserve exact raw whitespace and known empty
         const runId = (await readExecutionRuns(execution.databasePath, id)).rows[0].run_id;
         const response = await execution.request.get(`/prompt_manager/library/${value.positive_prompt_id}/history`);expect(response.ok()).toBe(true);
         expect((await response.json()).items.map(item => item.run_id)).toContain(runId);
+        if (kind !== 'PromptManager') continue;
+        // #460: an upper-case re-queue counts toward the same prompt and joins its history
+        const upper = await submit(execution.request, recipeGraph({ kind, text: `${kind.toUpperCase()} E\u0301` }));
+        await history(execution.request, upper);
+        const variant = (await capture(execution, upper)).contexts[0];
+        expect(variant.positive.raw).toBe(`${kind.toUpperCase()} E\u0301`);
+        expect(variant.positive_prompt_id, 'the case variant names the same prompt').toBe(value.positive_prompt_id);
+        const variantRun = (await readExecutionRuns(execution.databasePath, upper)).rows[0].run_id;
+        const scoped = await execution.request.get(`/prompt_manager/library/${value.positive_prompt_id}/history`);
+        expect((await scoped.json()).items.map(item => item.run_id)).toContain(variantRun);
     }
 });
 

tests/comfyui/fixtures/ledger_nodes.py

@@ -334,8 +334,15 @@ async def image_monitor(request: web.Request) -> web.Response:
 
 @PromptServer.instance.routes.get("/pm_execution_fixture/monitor")
 async def image_monitor_state(request: web.Request) -> web.Response:
-    """File names of keyless images waiting for their run's recorded outputs."""
+    """File names of images waiting for their run's recorded outputs: keyless
+    ones (``awaiting``) and foreign ones (``foreign``, #247)."""
     monitor = _plugin_module("utils.image_monitor")._monitor_instance
     handler = getattr(monitor, "handler", None)
     waiting = handler.awaiting_history() if handler is not None else []
-    return web.json_response({"awaiting": [os.path.basename(p) for p in waiting]})
+    foreign = handler.awaiting_foreign() if handler is not None else []
+    return web.json_response(
+        {
+            "awaiting": [os.path.basename(p) for p in waiting],
+            "foreign": [os.path.basename(p) for p in foreign],
+        }
+    )

tests/comfyui/image-attribution.spec.mjs

@@ -1,6 +1,7 @@
 // #342 on the real pinned host: the image monitor links an output only to a prompt
 // that provably produced it. An unknown node (PMHistoryCached) stands in for the
-// maintainer's LLM; KSampler is the fixture's deterministic stand-in.
+// maintainer's LLM; KSampler is the fixture's deterministic stand-in. #247: an output
+// of a workflow without a PromptManager node gets its own prompt.
 import { randomUUID } from 'node:crypto';
 import { execFile } from 'node:child_process';
 import { promisify } from 'node:util';
@@ -141,6 +142,68 @@ test('an output saved without metadata links once its recorded run names a store
     }
 });
 
+// #247: no PromptManager node at all, only ComfyUI's own text encoders, sampler and
+// saver (the fixture's inputs and decoder stand in for the model files).
+function foreignGraph({ text, seed = 1, prefix }) {
+    return {
+        '10': { class_type: 'PMHistoryInputs', inputs: {} },
+        '20': { class_type: 'CLIPTextEncode', inputs: { text, clip: ['10', 0] } },
+        '30': { class_type: 'CLIPTextEncode', inputs: { text: '', clip: ['10', 0] } },
+        '40': { class_type: 'KSampler', inputs: { model: ['10', 1], latent_image: ['10', 2], positive: ['20', 0],
+            negative: ['30', 0], seed, steps: 1, cfg: 1, sampler_name: 'euler', scheduler: 'normal', denoise: 1 } },
+        '60': { class_type: 'PMHistoryImages', inputs: { value: 64, batch: 1, latent: ['40', 0] } },
+        '70': { class_type: 'SaveImage', inputs: { images: ['60', 0], filename_prefix: prefix } },
+    };
+}
+
+test('an output of a workflow without a PromptManager node creates and links its prompt; a second run reuses it', async ({ execution }, testInfo) => {
+    const text = `foreign lighthouse at dusk ${testInfo.project.name} ${randomUUID()}`;
+    const first = await saved(execution.request, foreignGraph({ text, prefix: 'foreign/run' }));
+    await linkedTo(execution, first, text);
+    // Only the seed changes: a new file, the same text, so the same stored prompt.
+    const second = await saved(execution.request, foreignGraph({ text, seed: 2, prefix: 'foreign/run' }));
+    await linkedTo(execution, second, text);
+    const state = await readLinks(execution.databasePath, [first, second], [text]);
+    expect(state).toEqual({ links: { [first]: text, [second]: text }, stored: { [text]: true } });
+    // Decided from the graph at settle: nothing waits for the run's recorded outputs.
+    expect((await (await execution.request.get('/pm_execution_fixture/monitor')).json()).foreign).toEqual([]);
+});
+
+// CUJ 1: two branches saved under one prefix, so neither the graph nor the file name
+// names a file's branch; the run's recorded saver does, once its outputs are sealed.
+function sharedPrefixGraph(texts) {
+    const nodes = {
+        '10': { class_type: 'PMHistoryInputs', inputs: {} },
+        '30': { class_type: 'CLIPTextEncode', inputs: { text: '', clip: ['10', 0] } },
+    };
+    texts.forEach((text, index) => {
+        const base = (index + 1) * 100;
+        nodes[String(base)] = { class_type: 'CLIPTextEncode', inputs: { text, clip: ['10', 0] } };
+        nodes[String(base + 1)] = { class_type: 'KSampler', inputs: { model: ['10', 1], latent_image: ['10', 2],
+            positive: [String(base), 0], negative: ['30', 0], seed: index + 1, steps: 1, cfg: 1, sampler_name: 'euler',
+            scheduler: 'normal', denoise: 1 } };
+        nodes[String(base + 2)] = { class_type: 'PMHistoryImages', inputs: { value: 64, batch: 1, latent: [String(base + 1), 0] } };
+        nodes[String(base + 3)] = { class_type: 'SaveImage', inputs: { images: [String(base + 2), 0], filename_prefix: 'foreign/dataset' } };
+    });
+    return nodes;
+}
+
+test('each branch of a shared-prefix workflow without a PromptManager node links to its own prompt after the run', async ({ execution }, testInfo) => {
+    const tag = `${testInfo.project.name} ${randomUUID()}`;
+    const texts = [`dataset branch one harbor ${tag}`, `dataset branch two meadow ${tag}`];
+    const id = randomUUID();
+    const response = await execution.request.post('/prompt', { data: { prompt: sharedPrefixGraph(texts), prompt_id: id } });
+    expect(response.ok()).toBe(true);
+    let item;
+    await expect.poll(async () => {
+        item = (await (await execution.request.get(`/history/${id}`)).json())[id];
+        return item?.status?.messages?.some(([kind]) => kind === 'execution_success') || false;
+    }, { timeout: 15000 }).toBe(true);
+    const files = ['103', '203'].map(node => item.outputs[node].images[0].filename);
+    for (const [index, file] of files.entries()) await linkedTo(execution, file, texts[index]);
+    expect((await (await execution.request.get('/pm_execution_fixture/monitor')).json()).foreign).toEqual([]);
+});
+
 // #332 on the real host: the migration ran at startup and the image category routes
 // set, list and remove an own category on a generated output.
 test('an output gets its own image category and loses it again', async ({ execution }, testInfo) => {

tests/comfyui/inspect-sender.spec.mjs

@@ -36,11 +36,11 @@ async function watchHost(page,id,reset=true) {
 }
 const restore=page=>page.evaluate(async()=>{const {app}=await import('/scripts/app.js');Object.assign(app,window.inspectOriginals);});
 
-/** Chooses `file` on the Inspect page and opens the shared viewer on its local record. */
+/** Chooses `file` on the Inspect page, which opens the shared viewer on its local record (D155, #448). */
 async function inspect(workspace,{file,positive,shown}) {
     await workspace.locator('#local-file').setInputFiles(file());
     const open=workspace.locator('#workspace-main .heading-actions').getByRole('button',{name:'Open file metadata',exact:true});
-    await expect(open).toHaveAttribute('aria-disabled','false');await open.click();
+    await expect(open).toHaveAttribute('aria-disabled','false');
     const rail=workspace.locator('dialog#viewer-dialog[open] #viewer-details');
     await expect(rail.locator('.prompt-section.positive p.prompt-text')).toHaveText(positive);
     await expect(rail.locator('.prompt-section.negative p.prompt-text')).toHaveText(shown);

tests/fixtures/legacy_contract_snapshot.json

@@ -192,6 +192,7 @@
     "settings.monitoring_enabled",
     "settings.result_timeout",
     "settings.safety_backup",
+    "settings.save_foreign_runs",
     "settings.thumbnail_edge",
     "settings.thumbnail_format",
     "settings.thumbnail_quality",

tests/js/gallery-settings.test.js

@@ -79,6 +79,15 @@ test("the workspace keys have their own defaults and validation", () => {
     assert.equal(normalizeGallerySettings({ defaultViewMode: "masonry" }).defaultViewMode, "grid");
 });
 
+// #450 (D156): the viewer's default view, Centered unless Full screen was chosen.
+test("the viewer open mode defaults to centered and accepts only its two values", () => {
+    const { VIEWER_OPEN_MODES } = require("../../web/js/gallery-settings.js");
+    assert.deepEqual(VIEWER_OPEN_MODES, ["centered", "fullscreen"]);
+    assert.equal(DEFAULT_SETTINGS.viewerOpenMode, "centered");
+    assert.equal(normalizeGallerySettings({ viewerOpenMode: "fullscreen" }).viewerOpenMode, "fullscreen");
+    for (const value of ["full", true, 1, null]) assert.equal(normalizeGallerySettings({ viewerOpenMode: value }).viewerOpenMode, "centered");
+});
+
 // #330 S13: the legacy gallery's Page options edit only the page-only keys (PAGE_OPTION_KEYS); System › Appearance owns the rest, so a Save
 // or a Reset from the legacy page keeps every other stored key.
 function legacyGallery(stored, controls, extra = {}) {
@@ -112,7 +121,8 @@ const PAGE_OPTION_CONTROLS = () => ({
 
 test("a legacy gallery Save keeps the workspace keys and every other stored key", () => {
     const { manager, saved } = legacyGallery({ workspaceLayout: "masonry", workspaceDensity: "compact", someOtherTool: 1,
-        imageQuality: "high", showImageInfo: false, videoMute: false, thumbnailPromptDismissed: "sig", defaultViewMode: "list" },
+        imageQuality: "high", showImageInfo: false, videoMute: false, thumbnailPromptDismissed: "sig", defaultViewMode: "list",
+        viewerOpenMode: "fullscreen" },
         PAGE_OPTION_CONTROLS());
     manager.saveSettings();
     const result = saved();
@@ -123,6 +133,7 @@ test("a legacy gallery Save keeps the workspace keys and every other stored key"
     assert.equal(result.showImageInfo, false);
     assert.equal(result.videoMute, false);
     assert.equal(result.thumbnailPromptDismissed, "sig");
+    assert.equal(result.viewerOpenMode, "fullscreen");
     // Page options no longer edit the default view (S13 G5), so a stored one stays.
     assert.equal(result.defaultViewMode, "list");
     // The page-only keys come from the dialog.

tests/js/helpers/mounted-dom.mjs

@@ -10,6 +10,7 @@ export function mountedDom() {
         // As in the DOM, a string argument becomes a text node.
         static wrap(node){if(typeof node!=='string')return node;const text=new Node('#text');text.textContent=node;return text;}
         append(...nodes){for(const node of nodes.map(Node.wrap)){node.parent=this;this.children.push(node);}}
+        insertBefore(node,before){node.parent=this;const index=this.children.indexOf(before);this.children.splice(index<0?this.children.length:index,0,node);return node;}
         prepend(...nodes){nodes=nodes.map(Node.wrap);for(const node of nodes)node.parent=this;this.children.unshift(...nodes);}
         replaceChildren(...nodes){for(const child of this.children)child.parent=null;this.children=[];this._text='';this.append(...nodes);}
         remove(){if(this.parent)this.parent.children=this.parent.children.filter(c=>c!==this);this.parent=null;}

tests/js/workspace-library-state.test.js

@@ -104,6 +104,31 @@ test('archived variant reads use their own scope and cursor without changing the
  const active=[];const plain=createLibraryDetail({api:{get:async url=>{active.push(url);return {positive:item(1),variants:[]};}},onChange:()=>{}});await plain.load(1);assert.equal(new URL(active[0],'https://x').searchParams.has('archive'),false);
  assert.throws(()=>createLibraryDetail({api:{},onChange:()=>{},archive:'parents'}));
 });
+test('tag mode any and a folder are sent, echoed and validated (#444)',async()=>{
+ const {createLibraryStream,LIBRARY_TAG_MODES}=await source;assert.deepEqual(LIBRARY_TAG_MODES,['all','any','untagged']);
+ const calls=[];const pending=[{...page([item(1)]),tag_mode:'any',folder:'2026/Aug'},{...page([item(2)]),tag_mode:'all',folder:''},page([item(3)]),{...page([item(4)]),tag_mode:'all'}];
+ const s=createLibraryStream({api:{get:async(url)=>{calls.push(new URL(url,'https://x').searchParams);return pending.shift();}}});
+ assert.equal(await s.reset({tags:['a','b'],tagMode:'any',folder:'2026/Aug'}),1);
+ assert.equal(calls[0].get('tag_mode'),'any');assert.equal(calls[0].get('folder'),'2026/Aug');assert.deepEqual(calls[0].getAll('tag'),['a','b']);
+ // An older server that ignored them answers without the echo: an error, never the unfiltered shelf.
+ assert.equal(await s.reset({tags:['a','b'],tagMode:'any',folder:'2026/Aug'}),0);assert.ok(s.state.error instanceof TypeError);
+ // The defaults send neither, so they need no echo.
+ assert.equal(await s.reset({tagMode:'all',folder:''}),1);assert.equal(calls[2].has('tag_mode'),false);assert.equal(calls[2].has('folder'),false);
+ assert.equal(await s.reset({folder:'x'}),0);assert.ok(s.state.error instanceof TypeError,'a folder without its echo is refused');
+ await assert.rejects(s.reset({tagMode:'or'}),TypeError);await assert.rejects(s.reset({folder:3}),TypeError);
+ assert.equal(calls.length,4);s.dispose();
+});
+test('tag mode untagged is sent, echoed and never combined with tags (#444)',async()=>{
+ const {createLibraryStream}=await source;const calls=[];const pending=[{...page([item(1)]),tag_mode:'untagged'},page([item(2)]),page([item(3)])];
+ const s=createLibraryStream({api:{get:async(url)=>{calls.push(url);return pending.shift();}}});
+ assert.equal(await s.reset({tagMode:'untagged'}),1);assert.equal(new URL(calls[0],'https://x').searchParams.get('tag_mode'),'untagged');
+ // An older server that ignored tag_mode answers without the echo: an error, never the whole Library.
+ assert.equal(await s.reset({tagMode:'untagged'}),0);assert.ok(s.state.error instanceof TypeError);assert.deepEqual(s.state.items,[]);
+ // The default sends no tag_mode, so it needs no echo.
+ assert.equal(await s.reset({tagMode:'all'}),1);assert.equal(new URL(calls[2],'https://x').searchParams.has('tag_mode'),false);
+ await assert.rejects(s.reset({tagMode:'untagged',tags:['a']}),TypeError);await assert.rejects(s.reset({tagMode:'or'}),TypeError);
+ assert.equal(calls.length,3);s.dispose();
+});
 test('the stream keeps the first page\'s storage epoch and refuses a later page from another one (#437)',async()=>{
  const {createLibraryStream}=await source;const A='11111111-1111-4111-8111-111111111111',B='22222222-2222-4222-8222-222222222222';
  const pending=[{...page([item(1)],'next'),library_epoch:A},{...page([item(2)],'more'),library_epoch:A},{...page([item(3)]),library_epoch:B}];
@@ -120,3 +145,28 @@ test('the shelf no longer sorts by rating (D153)',async()=>{
  assert.throws(()=>createLibraryStream({api:{get:async()=>page([])},restore:{filters:{sort:'rating'},state:{}}}),/Invalid library sort/);
  for(const sort of ['newest','oldest','name']) assert.equal(await s.reset({sort}),0);s.dispose();
 });
+test('Refine › Folder nests each folder under its exact parent, by path segment (#444 live B1)',async()=>{
+ const {libraryFolderOptions}=await source;assert.equal(typeof libraryFolderOptions,'function');
+ const I='  ';
+ // The server's flat list sorts full paths as text, so "-" and "_" land before "/" (live report shapes).
+ const folders=['CCC','CCC/mabes9sans-dataset','CCC/mabes9sans-dataset-v1','CCC/mabes9sans-dataset/mabes9sans-dataset_IMG_GEN',
+  'FLUX','FLUX/2025-12-11','flux','flux/images','flux-klein','illustrious','illustrious-base','illustrious/2026-10-26','illustrious/training',
+  'illustrious/training/a','kiko','kiko-v6','kiko/char','kiko/XXX','QWEN','QWEN/x','Qwen','Qwen/y','Qwen-Training','Video','Video/v','video'].sort();
+ const options=libraryFolderOptions(folders);
+ assert.deepEqual(options[0],['','All output folders']);
+ assert.deepEqual(options.slice(1),[
+  ['CCC','CCC'],['CCC/mabes9sans-dataset',`${I}mabes9sans-dataset`],['CCC/mabes9sans-dataset/mabes9sans-dataset_IMG_GEN',`${I}${I}mabes9sans-dataset_IMG_GEN`],
+  ['CCC/mabes9sans-dataset-v1',`${I}mabes9sans-dataset-v1`],
+  ['FLUX','FLUX'],['FLUX/2025-12-11',`${I}2025-12-11`],['flux','flux'],['flux/images',`${I}images`],['flux-klein','flux-klein'],
+  ['illustrious','illustrious'],['illustrious/2026-10-26',`${I}2026-10-26`],['illustrious/training',`${I}training`],['illustrious/training/a',`${I}${I}a`],
+  ['illustrious-base','illustrious-base'],
+  ['kiko','kiko'],['kiko/char',`${I}char`],['kiko/XXX',`${I}XXX`],['kiko-v6','kiko-v6'],
+  ['QWEN','QWEN'],['QWEN/x',`${I}x`],['Qwen','Qwen'],['Qwen/y',`${I}y`],['Qwen-Training','Qwen-Training'],
+  ['Video','Video'],['Video/v',`${I}v`],['video','video'],
+ ]);
+ // Every parent precedes its children directly; a chosen folder missing from the list is kept, with its parents.
+ assert.deepEqual(libraryFolderOptions(['a'],'a/b/c').slice(1),[['a','a'],['a/b',`${I}b`],['a/b/c',`${I}${I}c`]]);
+ assert.deepEqual(libraryFolderOptions([],''),[['','All output folders']]);
+ // A POSIX folder whose name holds a backslash is one segment.
+ assert.deepEqual(libraryFolderOptions(['Metadata\\Klein_4b Wildcard Run']).slice(1),[['Metadata\\Klein_4b Wildcard Run','Metadata\\Klein_4b Wildcard Run']]);
+});

tests/js/workspace-media-viewer.test.js

@@ -188,6 +188,7 @@ test('an inline Viewer still loading its image is destroyed only after it has vi
     } finally { view.dispose(); parentScope.dispose(); f.restore(); Object.assign(globalThis, globals); }
 });
 
+const left0 = dialog => dialog.children[1].children[0];
 test('viewer renders the concept shell and toolbar order', async () => {
     const {reportedLocationAdapter:adapter} = await import('../../web/js/workspace/history-media.mjs');
     const f = mountedDom(), parentScope = createScope(), controller = borrowedController(), items = [occurrenceMedia('1',1,'c'),occurrenceMedia('2',0,'d')];
@@ -199,7 +200,14 @@ test('viewer renders the concept shell and toolbar order', async () => {
         assert.equal(dialog.attributes['aria-label'],undefined,'the title names the dialog');
         const [header,layout] = dialog.children;
         assert.equal(header.className,'dialog-header');
-        const [heading,close] = header.children, [title,subtitle] = heading.children;
+        const [heading,mode,close] = header.children, [title,subtitle] = heading.children;
+        // #450 D156: the view switch before ✕ names the view it switches to (Centered is the stored default).
+        assert.deepEqual([mode.tagName,mode.id,mode.className,mode.attributes['aria-label'],mode.title],['button','viewer-mode','icon-btn','Full screen view','Full screen view']);
+        mode.fire('click');
+        assert.equal(mode.attributes['aria-label'],'Centered view');
+        assert.ok(f.walk(left0(dialog),n => n.textContent === 'Hide details').length === 1,'full screen adds the details toggle');
+        mode.fire('click');
+        assert.equal(mode.attributes['aria-label'],'Full screen view');
         assert.deepEqual([heading.tagName,title.tagName,title.id,title.textContent],['div','h2','viewer-title','a.png']);
         assert.deepEqual([subtitle.tagName,subtitle.className,subtitle.textContent],['p','muted small','One viewer · image, prompts, generation data, workflow']);
         assert.deepEqual([close.tagName,close.id,close.textContent,close.attributes['aria-label']],['button','viewer-close','×','Close image viewer']);

tests/js/workspace-saved-views.test.js

@@ -5,7 +5,7 @@ const assert = require('node:assert/strict');
 const source = import('../../web/js/workspace/saved-views.mjs');
 
 const IMAGE_DEFAULTS = { q: '', category: '', sort: 'newest', source: 'all', folder: '', media: 'all', tags: [], favorites: false, collection: '' };
-const LIBRARY_DEFAULTS = { q: '', category: '', sort: 'newest', tags: [] };
+const LIBRARY_DEFAULTS = { q: '', category: '', sort: 'newest', tags: [], tag_mode: 'all', folder: '' };
 
 test('a saved state always holds every page key and keeps only values its page accepts', async () => {
     const { normalizeViewState, SAVED_VIEW_PAGES } = await source;
@@ -31,8 +31,8 @@ test('a saved state always holds every page key and keeps only values its page a
     assert.deepEqual(normalizeViewState('images', { ...full, q: 'x'.repeat(501), folder: 7, tags: Array(21).fill('t') }),
         { ...full, q: '', folder: '', tags: [] });
     assert.deepEqual(normalizeViewState('images', { ...full, category: 'bad\u0007', tags: [' '] }), { ...full, category: '', tags: [] });
-    // Library has no source, folder or media, and no Name Z–A.
-    assert.deepEqual(normalizeViewState('library', { ...full, sort: 'name-desc' }), { q: 'castle', category: 'Portraits', sort: 'newest', tags: ['moody', 'fog'] });
+    // Library has no source or media, and no Name Z–A; its folder (#444) is the Library's own.
+    assert.deepEqual(normalizeViewState('library', { ...full, sort: 'name-desc' }), { q: 'castle', category: 'Portraits', sort: 'newest', tags: ['moody', 'fog'], tag_mode: 'all', folder: 'a/b' });
     assert.deepEqual(normalizeViewState('library', { sort: 'oldest' }), { ...LIBRARY_DEFAULTS, sort: 'oldest' });
     // The returned tags are a copy.
     const tags = ['one'];
@@ -62,7 +62,7 @@ test('the listing keeps usable views in order and normalizes their state', async
         { id: 1, page: 'images', name: 'Videos', state: { media: 'video', source: '9' } },
     ] });
     assert.deepEqual(views, [
-        { id: 3, page: 'library', name: 'Neon', state: { q: 'neon', category: '', sort: 'name', tags: [] } },
+        { id: 3, page: 'library', name: 'Neon', state: { q: 'neon', category: '', sort: 'name', tags: [], tag_mode: 'all', folder: '' } },
         { id: 1, page: 'images', name: 'Videos', state: { ...IMAGE_DEFAULTS, media: 'video', source: '9' } },
     ]);
     assert.equal(parseSavedView(null), null);
@@ -84,8 +84,8 @@ test('page filters and saved states convert both ways', async () => {
     assert.equal(imagesViewState({ ...filters, collection: undefined }).collection, '');
     assert.equal(imagesViewState({ ...filters, collection: 7 }).collection, '7');
     assert.equal(imagesViewState({ ...filters, favorites: undefined }).favorites, false);
-    const shelf = { text: 'neon', category: 'Cities', sort: 'oldest', tags: ['night'] };
-    assert.deepEqual(libraryViewState(shelf), { q: 'neon', category: 'Cities', sort: 'oldest', tags: ['night'] });
+    const shelf = { text: 'neon', category: 'Cities', sort: 'oldest', tags: ['night'], tagMode: 'any', folder: '2026' };
+    assert.deepEqual(libraryViewState(shelf), { q: 'neon', category: 'Cities', sort: 'oldest', tags: ['night'], tag_mode: 'any', folder: '2026' });
     assert.deepEqual(libraryFilters(libraryViewState(shelf)), shelf);
 });
 
@@ -216,6 +216,27 @@ test('a state over the API 4 KB cap is refused before anything is sent (#249 M5)
     assert.deepEqual(api.calls, []);
 });
 
+test('a Library view keeps its tag match and folder; an old view reads the defaults (#444)', async () => {
+    const { normalizeViewState, libraryFilters } = await source;
+    assert.deepEqual(normalizeViewState('library', { tag_mode: 'any', folder: '2026/Aug' }), { ...LIBRARY_DEFAULTS, tag_mode: 'any', folder: '2026/Aug' });
+    assert.deepEqual(normalizeViewState('library', { tag_mode: 'or', folder: 'x'.repeat(1025) }), LIBRARY_DEFAULTS);
+    assert.deepEqual(libraryFilters({ q: 'a' }), { text: 'a', category: '', sort: 'newest', tags: [], tagMode: 'all', folder: '' });
+});
+
+test('a Library view keeps untagged, and untagged never travels with tags (#444)', async () => {
+    const { normalizeViewState, libraryFilters, libraryViewState } = await source;
+    assert.equal(normalizeViewState('library', { tag_mode: 'untagged' }).tag_mode, 'untagged');
+    assert.deepEqual(normalizeViewState('library', { tags: ['a'], tag_mode: 'untagged' }), { ...LIBRARY_DEFAULTS, tags: ['a'] });
+    const shelf = { text: '', category: '', sort: 'newest', tags: [], tagMode: 'untagged', folder: '' };
+    assert.equal(libraryViewState(shelf).tag_mode, 'untagged');
+    assert.deepEqual(libraryFilters(libraryViewState(shelf)), shelf);
+});
+
+test('a shelf without the #444 filters saves their defaults', async () => {
+    const { libraryViewState } = await source;
+    assert.deepEqual(libraryViewState({ text: '', category: '', sort: 'newest', tags: [] }), LIBRARY_DEFAULTS);
+});
+
 test('a view saved under the retired Highest rated sort opens on Recently added (D153)', async () => {
     const { normalizeViewState, imagesFilters, libraryFilters } = await source;
     assert.equal(normalizeViewState('images', { sort: 'rating' }).sort, 'newest');

tests/js/workspace-system-appearance.test.js

@@ -65,7 +65,7 @@ function fakeDOM() {
 
 const BROWSER = { workspaceLayout: 'grid', workspaceDensity: 'comfortable', autoLoadMetadata: true, cacheMetadata: true,
     imageQuality: 'medium', videoAutoplay: false, videoMute: true, videoLoop: true, showFilePaths: true, showImageInfo: true,
-    sidebarCollapsedByDefault: false };
+    sidebarCollapsedByDefault: false, viewerOpenMode: 'centered' };
 
 async function mounted({ host = { webui_display_mode: 'newtab', result_timeout: 5 }, browser = BROWSER, hostFails = false } = {}) {
     const { mountAppearance } = await appearanceModule;
@@ -80,9 +80,9 @@ async function mounted({ host = { webui_display_mode: 'newtab', result_timeout:
     });
     await drafts.load('browser');
     const scope = createScope();
-    const [surface] = mountAppearance(null, { scope, drafts });
-    const byId = id => dom.find(surface, node => node.dataset?.setting === id);
-    return { dom, drafts, scope, surface, byId, loadHost: async () => { hostValues = host; await drafts.load('host'); },
+    const [surface, viewer] = mountAppearance(null, { scope, drafts });
+    const byId = id => dom.find(surface, node => node.dataset?.setting === id) ?? dom.find(viewer, node => node.dataset?.setting === id);
+    return { dom, drafts, scope, surface, viewer, byId, loadHost: async () => { hostValues = host; await drafts.load('host'); },
         fixHost: () => { hostFails = false; } };
 }
 
@@ -98,13 +98,26 @@ test('the surface is the concept section with the concept rows in order, then th
         assert.deepEqual(titles, ['Default layout', 'Thumbnail size', 'Continuous browsing', 'Load metadata on selection',
             'Cache metadata in this browser', 'Image loading quality', 'Autoplay videos', 'Mute videos by default', 'Loop videos',
             'Show file paths', 'Open from ComfyUI', 'Node search auto-hide (seconds)', 'Show image details on cards',
-            'Collapse viewer details by default', 'Reset display defaults']);
+            'Reset display defaults']);
         const continuous = rows[2];
         assert.deepEqual(dom.tags(continuous.children[0]), ['h3', 'p']);
         assert.equal(continuous.children[1].className, 'chip blue');
         assert.equal(continuous.children[1].textContent, 'Infinite scroll');
         assert.equal(rows.at(-1).className, 'operation');
-        assert.equal(APPEARANCE_ROWS.filter(Array.isArray).length, 13);
+        assert.equal(APPEARANCE_ROWS.filter(Array.isArray).length, 12);
+    } finally { scope.dispose(); assert.equal(dom.listenerCount(), 0); dom.restore(); }
+});
+
+// #450 (D156, amends D97): a "Viewer" surface after "Browsing & display" holds the viewer's two defaults.
+test('the Viewer surface holds Open viewer in and the moved collapse row', async () => {
+    const { VIEWER_ROWS } = await appearanceModule;
+    const { dom, viewer, scope } = await mounted();
+    try {
+        assert.equal(viewer.className, 'surface');
+        assert.equal(viewer.children[0].textContent, 'Viewer');
+        assert.equal(viewer.children[1].textContent, 'How the image viewer opens; the button beside ✕ switches the current view.');
+        assert.deepEqual(viewer.children.slice(2).map(row => row.children[0].children[0].textContent), ['Open viewer in', 'Collapse viewer details by default']);
+        assert.deepEqual(VIEWER_ROWS.map(row => row[0]), ['viewer-mode', 'viewer-details']);
     } finally { scope.dispose(); assert.equal(dom.listenerCount(), 0); dom.restore(); }
 });
 
@@ -127,9 +140,11 @@ test('added rows reach their consumers: each control edits its stored key with t
             byId(id).checked = false; byId(id).dispatch('change');
         }
         for (const id of ['video-auto', 'viewer-details']) { byId(id).checked = true; byId(id).dispatch('change'); }
+        assert.deepEqual(byId('viewer-mode').children.map(option => [option.value, option.textContent]), [['centered', 'Centered'], ['fullscreen', 'Full screen']]);
+        byId('viewer-mode').value = 'fullscreen'; byId('viewer-mode').dispatch('change');
         assert.deepEqual(drafts.changes('browser'), { workspaceLayout: 'masonry', workspaceDensity: 'large', imageQuality: 'low',
             autoLoadMetadata: false, cacheMetadata: false, videoMute: false, videoLoop: false, showFilePaths: false, showImageInfo: false,
-            videoAutoplay: true, sidebarCollapsedByDefault: true });
+            videoAutoplay: true, sidebarCollapsedByDefault: true, viewerOpenMode: 'fullscreen' });
         assert.deepEqual(drafts.changes('host'), { webui_display_mode: 'popup' });
     } finally { scope.dispose(); dom.restore(); }
 });
@@ -154,7 +169,7 @@ test('the auto-hide timeout takes whole seconds from 0 to 300 only', async () =>
 
 test('reset restores defaults without touching other keys', async () => {
     const { DISPLAY_KEYS } = await appearanceModule;
-    const changed = { ...BROWSER, workspaceLayout: 'list', videoMute: false, showImageInfo: false, sidebarCollapsedByDefault: true };
+    const changed = { ...BROWSER, workspaceLayout: 'list', videoMute: false, showImageInfo: false, sidebarCollapsedByDefault: true, viewerOpenMode: 'fullscreen' };
     const { dom, drafts, scope, surface, loadHost } = await mounted({ browser: changed });
     try {
         await loadHost();
@@ -163,7 +178,7 @@ test('reset restores defaults without touching other keys', async () => {
         assert.equal(reset.textContent, 'Reset');
         reset.click();
         // #451: the default layout is Masonry.
-        assert.deepEqual(drafts.changes('browser'), { workspaceLayout: 'masonry', videoMute: true, showImageInfo: true, sidebarCollapsedByDefault: false });
+        assert.deepEqual(drafts.changes('browser'), { workspaceLayout: 'masonry', videoMute: true, showImageInfo: true, sidebarCollapsedByDefault: false, viewerOpenMode: 'centered' });
         // The host fields and every key outside the display defaults are left as they are.
         assert.deepEqual(drafts.changes('host'), { result_timeout: 9 });
         assert.deepEqual([...DISPLAY_KEYS].sort(), Object.keys(BROWSER).sort());

tests/js/workspace-system-sources.test.js

@@ -157,3 +157,20 @@ test('a valid configured source missing from the effective list is another spell
     assert.deepEqual(sourceRows({ sources, saved, current: saved }).rows.at(-1), { key: 'output/.', title: 'output',
         description: 'output/. · same folder as another source', scan: null, remove: 'output/.' });
 });
+
+test('D158: "Save prompts from other workflows" drafts save_foreign_runs, disabled while monitoring is off (#247)', async () => {
+    const { FOREIGN_RUNS, foreignRunsState } = await sourcesModule;
+    assert.deepEqual(FOREIGN_RUNS, { id: 'foreign-runs', key: 'save_foreign_runs', label: 'Save prompts from other workflows',
+        description: 'New outputs of workflows without a PromptManager node add their prompt to the library when it is certain.',
+        off: 'Applies while Watch for new outputs is on.' });
+    assert.deepEqual(foreignRunsState({ value: true, monitoring: true }), { disabled: false, checked: true, note: null });
+    assert.deepEqual(foreignRunsState({ value: false, monitoring: true }), { disabled: false, checked: false, note: null });
+    // Monitoring drafted or saved off: nothing is watched, so the row says why it is disabled and keeps its value.
+    assert.deepEqual(foreignRunsState({ value: true, monitoring: false }), { disabled: true, checked: true, note: FOREIGN_RUNS.off });
+    // Not read yet (or the host read failed): disabled without a value, the host rows' own note applies.
+    assert.deepEqual(foreignRunsState({ value: undefined, monitoring: undefined }), { disabled: true, checked: null, note: null });
+    assert.deepEqual(foreignRunsState({ value: undefined, monitoring: false }), { disabled: true, checked: null, note: null });
+    // Saved off while the running monitor still watches until the restart (D128 `stopping`): it can still be set.
+    assert.deepEqual(foreignRunsState({ value: true, monitoring: false, stopping: true }),
+        { disabled: false, checked: true, note: FOREIGN_RUNS.off });
+});

tests/js/workspace-vocabulary-state.test.js

@@ -461,6 +461,57 @@ test('after dispose an unconfirmed write invalidates and a failed write does not
     }
 });
 
+test('a tag stream sends its search and sort, replaces them on reset and needs the server to echo them (#444)', async () => {
+    const {createVocabularyStream, TAG_SORTS} = await source;
+    assert.deepEqual(TAG_SORTS.map(([value]) => value), ['created', 'name', 'name-desc', 'prompts', 'prompts-asc']);
+    assert.deepEqual(TAG_SORTS.map(([, label]) => label), ['Oldest first', 'Name A–Z', 'Name Z–A', 'Most prompts', 'Fewest prompts']);
+    const echo = (page, search, sort) => ({...page, search, sort, untagged_count: 4});
+    const {calls, api} = queued([
+        echo(tagPage([tag(1)], 'next'), 'cat', 'name'), echo(tagPage([tag(2)]), 'cat', 'name'),
+        echo(tagPage([tag(3)]), '', 'created'), tagPage([tag(4)]), echo(tagPage([tag(5)]), 'dog', 'created'),
+    ]);
+    const stream = createVocabularyStream({api, kind: 'tags', query: {search: ' cat ', sort: 'name'}});
+    await stream.loadMore();
+    assert.equal(params(calls[0].url).get('search'), 'cat');
+    assert.equal(params(calls[0].url).get('sort'), 'name');
+    assert.equal(stream.state.untaggedCount, 4);
+    await stream.loadMore();
+    assert.equal(params(calls[1].url).get('search'), 'cat');
+    assert.equal(params(calls[1].url).get('cursor'), 'next');
+    assert.deepEqual(stream.state.query, {search: 'cat', sort: 'name'});
+    // The default query sends neither, so an older server answers it unchanged.
+    await stream.reset({search: '', sort: 'created'});
+    assert.equal(params(calls[2].url).has('search'), false);
+    assert.equal(params(calls[2].url).has('sort'), false);
+    assert.deepEqual(stream.state.items.map(row => row.id), [3]);
+    // A server that ignored the search (no echo) is an error, never a wrong list.
+    await stream.reset({search: 'dog', sort: 'created'});
+    assert.deepEqual(stream.state.items, []);
+    assert.ok(stream.state.error instanceof TypeError);
+    assert.equal(stream.state.untaggedCount, 4, 'the last known untagged count survives a reset');
+    await stream.loadMore();
+    assert.deepEqual(stream.state.items.map(row => row.id), [5]);
+    assert.throws(() => stream.reset({search: 'x', sort: 'count'}), TypeError);
+    assert.throws(() => createVocabularyStream({api, kind: 'categories', query: {search: 'x'}}), TypeError);
+    assert.throws(() => createVocabularyStream({api, kind: 'tags', query: {search: 3}}), TypeError);
+    stream.dispose();
+});
+
+test('an invalid untagged_count rejects the page', async () => {
+    const {createVocabularyStream} = await source;
+    for (const untagged_count of [-1, 1.5, '2']) {
+        const stream = createVocabularyStream({api: queued([{...tagPage([tag(1)]), untagged_count}]).api, kind: 'tags'});
+        await stream.loadMore();
+        assert.ok(stream.state.error instanceof TypeError, String(untagged_count));
+        assert.equal(stream.state.untaggedCount, null);
+    }
+});
+
+test('untaggedBrowseFilters opens the Library on prompts without tags (#444)', async () => {
+    const {untaggedBrowseFilters} = await source;
+    assert.deepEqual(untaggedBrowseFilters(), {text: '', category: '', tags: [], sort: 'newest', tagMode: 'untagged'});
+});
+
 test('browseFilters hands Library one exact constraint', async () => {
     const {browseFilters} = await source;
     assert.deepEqual(browseFilters('tags', 'a'), {text: '', category: '', tags: ['a'], sort: 'newest'});

tests/test_api_core.py

@@ -684,6 +684,132 @@ class TestStaticRoutes(CoreRoutesTestCase):
         self.assertEqual(resp.status, 403)
 
 
+class TestRevalidatedCaching(CoreRoutesTestCase):
+    """Every served page and static file is revalidated on reload (#478).
+
+    ``Cache-Control: no-cache`` makes the browser ask before it reuses a copy,
+    and the ETag lets an unchanged file answer 304 without a body, so an update
+    reaches the browser on a normal reload and an unchanged reload stays cheap.
+    """
+
+    STATIC_PATHS = (
+        "/prompt_manager/js/workspace/index.mjs",
+        "/prompt_manager/js/admin.js",
+        "/prompt_manager/js/inspection-semantics.json",
+        "/prompt_manager/lib/viewerjs/viewer.min.css",
+    )
+
+    async def _assert_revalidated(self, path):
+        resp = await self.client.request("GET", path)
+        self.assertEqual(resp.status, 200, path)
+        self.assertEqual(resp.headers.get("Cache-Control"), "no-cache", path)
+        etag = resp.headers.get("ETag")
+        self.assertTrue(etag, path)
+        body = await resp.read()
+        self.assertTrue(body, path)
+
+        again = await self.client.request("GET", path, headers={"If-None-Match": etag})
+        self.assertEqual(again.status, 304, path)
+        self.assertEqual(again.headers.get("ETag"), etag, path)
+        self.assertEqual(again.headers.get("Cache-Control"), "no-cache", path)
+        self.assertEqual(await again.read(), b"", path)
+
+        stale = await self.client.request(
+            "GET", path, headers={"If-None-Match": 'W/"stale"'}
+        )
+        self.assertEqual(stale.status, 200, path)
+        self.assertEqual(await stale.read(), body, path)
+        return etag
+
+    async def test_static_files_are_revalidated(self):
+        for path in self.STATIC_PATHS:
+            await self._assert_revalidated(path)
+
+    async def test_pages_are_revalidated(self):
+        for path in UI_PATHS:
+            await self._assert_revalidated(path)
+
+    async def test_etag_matches_in_a_list_and_as_wildcard(self):
+        path = "/prompt_manager/js/workspace/index.mjs"
+        etag = (await self.client.request("GET", path)).headers["ETag"]
+        for header in (f'W/"other", {etag}', etag.removeprefix("W/"), "*"):
+            resp = await self.client.request(
+                "GET", path, headers={"If-None-Match": header}
+            )
+            self.assertEqual(resp.status, 304, header)
+
+    async def test_changed_file_gets_a_new_etag_and_its_new_body(self):
+        web_dir = self._fake_root()
+        module = web_dir / "js" / "m.mjs"
+        module.write_text("export const v = 1;")
+        path = "/prompt_manager/js/m.mjs"
+        first = await self.client.request("GET", path)
+        etag = first.headers["ETag"]
+        # Same size and, on coarse clocks, the same mtime: only the bytes differ.
+        module.write_text("export const v = 2;")
+        resp = await self.client.request("GET", path, headers={"If-None-Match": etag})
+        self.assertEqual(resp.status, 200)
+        self.assertNotEqual(resp.headers["ETag"], etag)
+        self.assertEqual(await resp.text(), "export const v = 2;")
+
+    async def test_gzipped_and_plain_bodies_share_a_weak_etag(self):
+        """The gzip middleware changes the bytes, so the tag must be weak."""
+        resp = await self.client.request("GET", "/prompt_manager/js/admin.js")
+        self.assertTrue(resp.headers["ETag"].startswith('W/"'))
+
+    async def test_errors_carry_no_validator(self):
+        resp = await self.client.request("GET", "/prompt_manager/js/nope.js")
+        self.assertEqual(resp.status, 404)
+        self.assertNotIn("ETag", resp.headers)
+
+
+@web.middleware
+async def _comfyui_cache_control(request, handler):
+    """ComfyUI's own middleware (middleware/cache_middleware.py), JS/CSS branch.
+
+    It only fills a missing Cache-Control, and ``.mjs`` is not ``.js``: that is
+    why the workspace modules went out with no header at all before #478.
+    """
+    response = await handler(request)
+    if request.path.endswith(".js") or request.path.endswith(".css"):
+        response.headers.setdefault("Cache-Control", "no-store")
+    return response
+
+
+class TestRevalidatedCachingBehindMiddlewares(CoreRoutesTestCase):
+    """The host's middlewares keep the header and the 304 intact."""
+
+    async def get_application(self):
+        app = await super().get_application()
+        app.middlewares.append(_comfyui_cache_control)
+        app.middlewares.append(api_module._gzip_middleware)
+        return app
+
+    async def test_gzipped_files_keep_no_cache_and_answer_304(self):
+        for path in (
+            "/prompt_manager/js/workspace/index.mjs",
+            "/prompt_manager/js/admin.js",
+            "/prompt_manager/lib/viewerjs/viewer.min.css",
+            "/prompt_manager/workspace",
+        ):
+            resp = await self.client.request(
+                "GET", path, headers={"Accept-Encoding": "gzip"}
+            )
+            self.assertEqual(resp.status, 200, path)
+            self.assertEqual(resp.headers.get("Content-Encoding"), "gzip", path)
+            self.assertEqual(resp.headers["Cache-Control"], "no-cache", path)
+            again = await self.client.request(
+                "GET",
+                path,
+                headers={
+                    "Accept-Encoding": "gzip",
+                    "If-None-Match": resp.headers["ETag"],
+                },
+            )
+            self.assertEqual(again.status, 304, path)
+            self.assertEqual(again.headers["Cache-Control"], "no-cache", path)
+
+
 # ── instance helpers ──────────────────────────────────────────────────
 
 

tests/test_api_diagnostics_report.py

@@ -85,6 +85,7 @@ CONFIG = {
         "max_image_age_days",
         "max_concurrent_processing",
         "metadata_extraction_timeout",
+        "save_foreign_runs",
     },
     "integrations": {"lora_manager_enabled", "has_civitai_api_key"},
 }

tests/test_api_history.py

@@ -18,6 +18,8 @@ from unittest.mock import patch
 
 from aiohttp import FormData, web
 
+from utils.hashing import generate_prompt_hash
+
 try:
     from test_api_images import ImageAPITestCase
 except ImportError:
@@ -191,7 +193,11 @@ class HistoryAPITests(ImageAPITestCase):
             context["evidence"]["positive"]["raw"]["provenance"], "recorded"
         )
         with self.api.db.model.get_connection() as conn:
-            conn.execute("UPDATE prompts SET text='edited' WHERE id=?", (positive,))
+            # An edit changes the text and its dedup hash together (#460)
+            conn.execute(
+                "UPDATE prompts SET text='edited',hash=? WHERE id=?",
+                (generate_prompt_hash("edited"), positive),
+            )
         _, changed = await self.detail(start.run_id)
         self.assertEqual(changed["contexts"][0]["positive"], context["positive"])
         self.assertFalse(changed["contexts"][0]["navigation"]["positive_available"])

tests/test_api_history_outputs.py

@@ -1015,6 +1015,122 @@ class HistoryOutputAPITests(ImageAPITestCase):
             response = await self.current(descriptor["inspection_url"])
             self.assert_error(response, await response.json(), 422, "response-limit")
 
+    async def tag_image(self, relative, tags):
+        """Save *tags* on Images root 0's *relative* file through the Tag routes.
+
+        The real observe and save routes store the locator the Tag dialog
+        stores (``media_capture._resolve``), so History must find it (#457).
+        """
+        self.addCleanup(self.api.close_annotation_captures)
+        roots = self.api._get_all_output_dirs()
+        target = {
+            "root_index": 0,
+            "root_scope": self.api._gallery_root_scope(roots[0], 0),
+            "path": relative,
+        }
+        response = await self.client.post(
+            "/prompt_manager/image-annotations/observe", json={"target": target}
+        )
+        observed = await response.json()
+        self.assertEqual(response.status, 200, observed)
+        token = observed["capture"]["token"]
+        try:
+            response = await self.client.put(
+                "/prompt_manager/image-annotations/tags",
+                json={
+                    "target": target,
+                    "capture_token": token,
+                    "sha256": observed["capture"]["sha256"],
+                    "expected_epoch": observed["library_epoch"],
+                    "expected_tag_revision": observed["tag_revision"],
+                    "tags": list(tags),
+                },
+            )
+            saved = await response.json()
+            self.assertEqual(response.status, 200, saved)
+        finally:
+            await self.client.delete(
+                f"/prompt_manager/image-annotations/capture/{token}"
+            )
+        return saved["tags"]
+
+    async def test_current_inspection_carries_the_files_image_tags(self):
+        """#457: History reads the file's own image tags, as Images does (S1)."""
+        sub = self.output_dir / "batch"
+        sub.mkdir()
+        tagged = png(sub / "tagged.png", prompt=graph("Tagged file"))
+        png(self.output_dir / "plain.png", prompt=graph("Plain file"))
+        self.assertEqual(
+            await self.tag_image("batch/tagged.png", ["dunes", "warm_light"]),
+            ["dunes", "warm_light"],
+        )
+        start = self.capture(
+            [
+                self.ref("1", subfolder="batch", filename="tagged.png"),
+                self.ref("2", filename="plain.png"),
+            ]
+        )
+        _, page = await self.outputs(start)
+        urls = [
+            item["current_candidate"]["media"]["inspection_url"]
+            for item in page["items"]
+        ]
+        bodies = []
+        for url in urls:
+            response = await self.current(url)
+            body = await response.json()
+            self.assertEqual(response.status, 200, body)
+            bodies.append(body)
+        self.assertEqual(
+            bodies[0].get("image_tags"),
+            {"state": "current", "tags": ["dunes", "warm_light"]},
+        )
+        # Image tags are the file's own search metadata, never prompt tags.
+        self.assertEqual(bodies[0]["prompts"]["positive"]["text"], "Tagged file")
+        self.assertEqual(bodies[1].get("image_tags"), {"state": "none", "tags": []})
+        # New bytes at the same path: the tags belong to the old byte record.
+        stat = tagged.stat()
+        os.utime(tagged, ns=(stat.st_atime_ns, stat.st_mtime_ns + 10**9))
+        response = await self.current(urls[0])
+        body = await response.json()
+        self.assertEqual(response.status, 200, body)
+        self.assertEqual(body["image_tags"], {"state": "none", "tags": []})
+
+    async def test_current_inspection_finds_image_tags_under_a_symlinked_output(self):
+        """An output folder reached through a link (another drive) still matches."""
+        link = self.comfy_dir / "linked_output"
+        self._make_symlink(self.output_dir, link, target_is_directory=True)
+        self.api._get_all_output_dirs = lambda: [link]
+        self.host_output = str(link)
+        png(self.output_dir / "image.png", prompt=graph("Linked folder"))
+        self.assertEqual(await self.tag_image("image.png", ["dunes"]), ["dunes"])
+        start = self.capture([self.ref()])
+        descriptor = await self.media(start)
+        response = await self.current(descriptor["inspection_url"])
+        body = await response.json()
+        self.assertEqual(response.status, 200, body)
+        self.assertEqual(body["image_tags"], {"state": "current", "tags": ["dunes"]})
+
+    async def test_current_inspection_image_tag_failure_is_unavailable(self):
+        png(self.output_dir / "image.png", prompt=graph("File evidence"))
+        await self.tag_image("image.png", ["dunes"])
+        start = self.capture([self.ref()])
+        descriptor = await self.media(start)
+        with (
+            self.assertLogs(self.api.logger, "WARNING") as logs,
+            patch.object(
+                type(self.api.db),
+                "current_image_tags",
+                side_effect=sqlite3.OperationalError("locked"),
+            ),
+        ):
+            response = await self.current(descriptor["inspection_url"])
+            body = await response.json()
+        self.assertEqual(response.status, 200, body)
+        self.assertEqual(body["image_tags"], {"state": "unavailable", "tags": []})
+        self.assertEqual(body["prompts"]["positive"]["text"], "File evidence")
+        self.assertNotIn(str(self.output_dir), "\n".join(logs.output))
+
     async def test_optional_candidate_failure_preserves_every_occurrence_without_urls(
         self,
     ):

tests/test_api_images.py

@@ -35,6 +35,7 @@ from aiohttp import web  # noqa: E402
 from aiohttp.test_utils import AioHTTPTestCase  # noqa: E402
 from PIL import Image  # noqa: E402
 
+from database.models import PromptModel  # noqa: E402
 from database.operations import PromptDatabase  # noqa: E402
 import py.api.images as images_module
 from py.api import gallery_listing  # noqa: E402
@@ -92,14 +93,41 @@ class ImageAPITestCase(AioHTTPTestCase):
         self._anchor_patch.stop()
         # Windows refuses to unlink a database that still has open handles.
         self.api.db.close_all()
-        for path in (
-            self._temp_db.name,
-            self._temp_db.name + "-wal",
-            self._temp_db.name + "-shm",
-        ):
-            if os.path.exists(path):
-                os.unlink(path)
-        self._output_tmp.cleanup()
+        try:
+            leaked = self._close_leaked_connections()
+        finally:
+            for path in (
+                self._temp_db.name,
+                self._temp_db.name + "-wal",
+                self._temp_db.name + "-shm",
+            ):
+                if os.path.exists(path):
+                    os.unlink(path)
+            self._output_tmp.cleanup()
+        if leaked:
+            raise AssertionError(
+                f"{leaked} connection(s) to the test database were still open at "
+                "teardown; Windows cannot unlink it (WinError 32). Close every "
+                "extra PromptDatabase before tearDownAsync, not in a cleanup."
+            )
+
+    def _close_leaked_connections(self):
+        """Close and count connections other models still hold on the temp DB.
+
+        Linux unlinks an open SQLite file, Windows does not; counting here makes
+        a handle left open past teardown fail on every OS (#475).
+        """
+        key = os.path.normcase(os.path.realpath(self._temp_db.name))
+        with PromptModel._instances_lock:
+            models = list(PromptModel._instances)
+        leaked = 0
+        for model in models:
+            if model._schema_key() != key:
+                continue
+            with model._conn_lock:
+                leaked += len(model._connections)
+            model.close_all()
+        return leaked
 
     # -- helpers ---------------------------------------------------------
 

tests/test_api_library.py

@@ -60,6 +60,32 @@ class TestLibraryAPI(ImageAPITestCase):
             self.assertEqual(response.status, 400)
             self.assertNotIn(str(self.comfy_dir), json.dumps(result))
 
+    async def test_untagged_tag_mode_lists_parents_without_tags(self):
+        self._save_prompt("plain one")
+        self._save_prompt("plain two")
+        self._save_prompt("tagged", tags=["t"])
+        self.api.db.save_prompt_pair("pair tagged", "noise", tags=["p"])
+        _, first = await self.get(limit=1, tag_mode="untagged")
+        self.assertEqual(first["total"], 2)
+        self.assertEqual(first["tag_mode"], "untagged")
+        _, second = await self.get(
+            limit=1, tag_mode="untagged", cursor=first["next_cursor"]
+        )
+        self.assertEqual(
+            sorted([first["items"][0]["text"], second["items"][0]["text"]]),
+            ["plain one", "plain two"],
+        )
+        _, default = await self.get(tag_mode="all")
+        self.assertEqual(default["total"], 4)
+        for params in (
+            {"tag_mode": "untagged", "tag": "t"},
+            {"tag_mode": "sql"},
+            {"cursor": first["next_cursor"]},
+        ):
+            response, result = await self.get(limit=1, **params)
+            self.assertEqual(response.status, 400, params)
+            self.assertEqual(result["success"], False)
+
     async def test_outside_missing_and_unsupported_media_do_not_leak_or_scan(self):
         positive = self._save_prompt("Cat")
         outside = make_png(self.comfy_dir / "outside.png")
@@ -97,6 +123,72 @@ class TestLibraryAPI(ImageAPITestCase):
             },
         )
 
+    async def test_folder_filter_lists_and_matches_prompts_by_output_folder(self):
+        deep = self._save_prompt("deep")
+        top = self._save_prompt("top")
+        other = self._save_prompt("other")
+        self._save_prompt("no images")
+        for prompt, parts in ((deep, ("a", "b")), (top, ("a",)), (other, ("ab",))):
+            path = make_png(self.output_dir.joinpath(*parts, f"{prompt}.png"))
+            self.api.db.link_image_to_prompt(prompt, str(path))
+        response, options = await self.get("/options", include="folders")
+        self.assertEqual(response.status, 200)
+        self.assertEqual(options["folders"], ["a", "a/b", "ab"])
+        _, plain = await self.get("/options")
+        self.assertNotIn("folders", plain)
+        _, first = await self.get(folder="a", limit=1)
+        self.assertEqual((first["total"], first["folder"]), (2, "a"))
+        # Outer slashes are dropped (a backslash separates only on Windows).
+        outer = "\\a\\" if os.name == "nt" else "/a/"
+        _, second = await self.get(folder=outer, limit=1, cursor=first["next_cursor"])
+        self.assertEqual(
+            sorted([first["items"][0]["text"], second["items"][0]["text"]]),
+            ["deep", "top"],
+        )
+        _, all_rows = await self.get()
+        self.assertEqual((all_rows["total"], all_rows["folder"]), (4, ""))
+        for params in (
+            {"folder": "a/../ab"},
+            *([{"folder": "C:/x"}] if os.name == "nt" else []),
+            {"folder": "b", "cursor": first["next_cursor"]},
+            {"cursor": first["next_cursor"]},
+        ):
+            response, result = await self.get(limit=1, **params)
+            self.assertEqual(response.status, 400, params)
+            self.assertEqual(result["success"], False)
+        response, _ = await self.get("/options", include="tags")
+        self.assertEqual(response.status, 400)
+
+    @unittest.skipIf(os.name == "nt", "a backslash is a separator on Windows")
+    async def test_a_posix_folder_name_with_a_backslash_filters_its_prompts(self):
+        prompt = self._save_prompt("klein")
+        name = "Metadata\\Klein_4b Wildcard Run"
+        path = make_png(self.output_dir / name / "x.png")
+        self.api.db.link_image_to_prompt(prompt, str(path))
+        _, options = await self.get("/options", include="folders")
+        self.assertIn(name, options["folders"])
+        _, result = await self.get(folder=name)
+        self.assertEqual((result["total"], result["folder"]), (1, name))
+
+    async def test_any_tag_mode_is_sent_through_and_echoed(self):
+        self._save_prompt("a only", tags=["a"])
+        self._save_prompt("b only", tags=["b"])
+        self._save_prompt("neither")
+        _, result = await self.get(tag=["a", "b"], tag_mode="any")
+        self.assertEqual(result["tag_mode"], "any")
+        self.assertEqual(
+            sorted(i["text"] for i in result["items"]), ["a only", "b only"]
+        )
+        _, both = await self.get(tag=["a", "b"])
+        self.assertEqual((both["total"], both["tag_mode"]), (0, "all"))
+        _, first = await self.get(tag=["a", "b"], tag_mode="any", limit=1)
+        response, _ = await self.get(
+            tag=["a", "b"], limit=1, cursor=first["next_cursor"]
+        )
+        self.assertEqual(response.status, 400)
+        response, _ = await self.get(tag_mode="or")
+        self.assertEqual(response.status, 400)
+
     async def test_variant_cursor_is_positive_scoped_and_required_limits_are_bounded(
         self,
     ):

tests/test_api_vocabulary.py

@@ -35,7 +35,17 @@ class TestVocabularyAPI(ImageAPITestCase):
         self.assertEqual(response.status, 200)
         self.assertEqual(
             set(first),
-            {"success", "library_epoch", "tags", "total", "has_more", "next_cursor"},
+            {
+                "success",
+                "library_epoch",
+                "tags",
+                "total",
+                "untagged_count",
+                "search",
+                "sort",
+                "has_more",
+                "next_cursor",
+            },
         )
         self.assertEqual(first["library_epoch"], epoch)
         self.assertEqual(first["total"], 3)
@@ -111,6 +121,41 @@ class TestVocabularyAPI(ImageAPITestCase):
         response, _ = await self.get("tags", limit=1, cursor=fresh["next_cursor"])
         self.assertEqual(response.status, 200)
 
+    async def test_tag_search_and_sort_page_with_scoped_cursors(self):
+        self._save_prompt("One", tags=["beta", "alpha"])
+        self._save_prompt("Two", tags=["beta"])
+        self._save_prompt("Plain")
+        self.add_unused("gamma")
+        _, first = await self.get("tags", limit=1, sort="name", search="A")
+        self.assertEqual([row["name"] for row in first["tags"]], ["alpha"])
+        self.assertEqual(first["total"], 3)
+        self.assertEqual(first["untagged_count"], 1)
+        self.assertEqual((first["search"], first["sort"]), ("A", "name"))
+        _, second = await self.get(
+            "tags", limit=1, sort="name", search="A", cursor=first["next_cursor"]
+        )
+        self.assertEqual([row["name"] for row in second["tags"]], ["beta"])
+        _, ranked = await self.get("tags", sort="prompts")
+        self.assertEqual(
+            [(row["name"], row["prompt_count"]) for row in ranked["tags"]],
+            [("beta", 2), ("alpha", 1), ("gamma", 0)],
+        )
+        # A cursor belongs to one search and one sort.
+        for params in (
+            {"sort": "name-desc", "search": "A"},
+            {"sort": "name", "search": "B"},
+            {"sort": "created", "search": "A"},
+        ):
+            response, result = await self.get(
+                "tags", limit=1, cursor=first["next_cursor"], **params
+            )
+            self.assertEqual(response.status, 400, params)
+            self.assertIn("refresh", result["error"])
+        for params in ({"sort": "count"}, {"search": "x" * 201}):
+            response, result = await self.get("tags", **params)
+            self.assertEqual(response.status, 400, params)
+            self.assertEqual(result["success"], False)
+
     async def test_read_failures_keep_the_read_envelope(self):
         with patch.object(
             self.api.db,

tests/test_api_workspace_restart.py

@@ -38,11 +38,25 @@ def _payload(token):
 
 class TestRestartKeepsTheOpenPage(ImageAPITestCase):
     async def get_application(self):
+        self._restarts = []
         app = await super().get_application()
         for index in range(3):
             make_png(self.output_dir / f"image_{index}.png")
         return app
 
+    async def tearDownAsync(self):
+        # Before the base teardown unlinks the database: cleanups run after
+        # tearDownAsync, and Windows cannot unlink a file still open (#475).
+        errors = []
+        while self._restarts:
+            try:
+                await self._restarts.pop()()
+            except Exception as error:  # every closer runs; the first error wins
+                errors.append(error)
+        await super().tearDownAsync()
+        if errors:
+            raise errors[0]
+
     async def first_page(self, client=None, **params):
         response = await (client or self.client).get(
             LISTING, params={"mode": "workspace", "limit": "1", **params}
@@ -67,7 +81,7 @@ class TestRestartKeepsTheOpenPage(ImageAPITestCase):
             api.close_auto_thumbnails()
             api.db.close_all()
 
-        self.addAsyncCleanup(close)
+        self._restarts.append(close)
         return api, client
 
     async def test_a_restart_keeps_every_held_scope_and_cursor_valid(self):
@@ -152,8 +166,13 @@ class TestRestartKeepsTheOpenPage(ImageAPITestCase):
         other_db = self.comfy_dir / "elsewhere" / "prompts.db"
         other_db.parent.mkdir()
         api = PromptManagerAPI(db=PromptDatabase(str(other_db)))
-        self.addCleanup(api.db.close_all)
-        self.addCleanup(api.close_auto_thumbnails)
+
+        async def close():
+            api.close_auto_thumbnails()
+            api.db.close_all()
+
+        # Closed before teardown removes the folder that holds its database.
+        self._restarts.append(close)
         self.assertNotEqual(
             api._gallery_root_scope(self.output_dir, 0), page["images"][0]["root_scope"]
         )

tests/test_config.py

@@ -59,6 +59,28 @@ class TestGalleryConfig(unittest.TestCase):
         self.assertFalse(GalleryConfig.MONITORING_ENABLED)
         self.assertEqual(GalleryConfig.PROCESSING_DELAY, 5.0)
 
+    def test_save_foreign_runs_defaults_on_and_only_booleans_apply(self):
+        """#247: gallery.monitoring.save_foreign_runs, default true."""
+        self.assertIs(
+            GalleryConfig.get_config()["monitoring"]["save_foreign_runs"], True
+        )
+        GalleryConfig.update_config({"monitoring": {"save_foreign_runs": False}})
+        self.assertIs(GalleryConfig.SAVE_FOREIGN_RUNS, False)
+        for junk in ("true", 1, None):
+            with self.subTest(value=junk):
+                with self.assertLogs("prompt_manager.config", "WARNING") as logs:
+                    GalleryConfig.update_config(
+                        {"monitoring": {"save_foreign_runs": junk}}
+                    )
+                self.assertIn(
+                    "Ignoring gallery monitoring save_foreign_runs: expected true "
+                    "or false",
+                    "\n".join(logs.output),
+                )
+                self.assertIs(GalleryConfig.SAVE_FOREIGN_RUNS, False)
+        GalleryConfig.update_config({"monitoring": {"save_foreign_runs": True}})
+        self.assertIs(GalleryConfig.SAVE_FOREIGN_RUNS, True)
+
     def test_update_tracking(self):
         GalleryConfig.update_config({"tracking": {"prompt_timeout": 300}})
         self.assertEqual(GalleryConfig.PROMPT_TIMEOUT, 300)

tests/test_execution_context_database.py

@@ -15,6 +15,7 @@ from database import execution_contexts
 from database.execution_history import HistoryEpochChanged
 from database.execution_context_schema import context_schema_error
 from database.execution_contexts import LIVE_HINT_PROMPT_IDS_SQL
+from utils.hashing import generate_prompt_hash
 from database.execution_history_schema import rotate_history_epoch
 from utils.execution_contexts import (
     ContextBundle,
@@ -466,7 +467,11 @@ class ContextDatabaseTests(unittest.TestCase):
         with conn:
             conn.execute("DELETE FROM prompts WHERE id=3")
         self.assertEqual(assert_same(), {1, 2, 4})
+        # A case-only edit keeps the prompt's hash, so the run still counts
+        # toward it (#460); another text drops the hint.
         db.update_prompt_text(4, "Other")
+        self.assertEqual(assert_same(), {1, 2, 4})
+        db.update_prompt_text(4, "another")
         self.assertEqual(assert_same(), {1, 2})
         with conn:
             conn.execute("INSERT INTO prompts(id,text) VALUES(5,'none')")
@@ -522,6 +527,83 @@ class ContextDatabaseTests(unittest.TestCase):
         live = {row[0] for row in conn.execute(LIVE_HINT_PROMPT_IDS_SQL)}
         self.assertEqual(live, {1, 2})
 
+    def test_hints_follow_the_hash_for_case_variant_raw_text(self):
+        # #460: the run names the prompt the Library counted it toward.
+        db = self.database()
+        first = db.save_prompt(
+            "\u00e9t\u00e9 positive",
+            prompt_hash=generate_prompt_hash("\u00e9t\u00e9 positive"),
+        )
+        second = db.save_prompt(
+            "negative \u03c3", prompt_hash=generate_prompt_hash("negative \u03c3")
+        )
+        with db.model.get_connection() as conn:
+            conn.execute(
+                "INSERT INTO prompt_pairs(id,positive_prompt_id,negative_prompt_id) "
+                "VALUES(7,?,?)",
+                (first, second),
+            )
+        start = self.capture(
+            db, 0, bundle("NEGATIVE \u03a3", positive="\u00c9T\u00c9 Positive")
+        )
+        context = db.get_execution_contexts(start.run_id)["contexts"][0]
+        self.assertEqual(
+            (
+                context["positive_prompt_id"],
+                context["negative_prompt_id"],
+                context["pair_id"],
+            ),
+            (first, second, 7),
+        )
+        conn = db.model.get_connection()
+        live = {row[0] for row in conn.execute(LIVE_HINT_PROMPT_IDS_SQL)}
+        self.assertEqual(live, {first, second})
+        # A prompt that only shares SQLite's ASCII lower() is not a match
+        other = db.save_prompt("other", prompt_hash=generate_prompt_hash("other"))
+        start = self.capture(db, 1, bundle("x", positive="\u00d6THER"))
+        context = db.get_execution_contexts(start.run_id)["contexts"][0]
+        self.assertIsNone(context["positive_prompt_id"])
+        self.assertNotIn(
+            other, {row[0] for row in conn.execute(LIVE_HINT_PROMPT_IDS_SQL)}
+        )
+
+    def test_hash_is_only_a_fallback_so_a_text_hint_stays_unambiguous(self):
+        # #460 review: a hash match must never make a #438 text hint ambiguous,
+        # or its prompt would lose the recorded-run prune protection.
+        db = self.database()
+        conn = db.model.get_connection()
+        with conn:
+            # A legacy row without a hash, and a case variant with the hash
+            conn.execute("INSERT INTO prompts(id,text) VALUES(1,'a lighthouse')")
+            conn.execute(
+                "INSERT INTO prompts(id,text,hash) VALUES(2,'A Lighthouse',?)",
+                (generate_prompt_hash("a lighthouse"),),
+            )
+        exact = self.capture(db, 0, bundle("x", positive="a lighthouse"))
+        variant = self.capture(db, 1, bundle("x", positive="A LIGHTHOUSE"))
+        hints = [
+            db.get_execution_contexts(start.run_id)["contexts"][0]["positive_prompt_id"]
+            for start in (exact, variant)
+        ]
+        self.assertEqual(hints, [1, 2])
+        live = {row[0] for row in conn.execute(LIVE_HINT_PROMPT_IDS_SQL)}
+        self.assertEqual(live, {1, 2})
+        self.assertEqual(db.prune_orphaned_prompts(), 0)
+
+    def test_live_hint_prompt_lookups_use_indexes(self):
+        db = self.database()
+        conn = db.model.get_connection()
+        plan = [
+            row[3]
+            for row in conn.execute("EXPLAIN QUERY PLAN " + LIVE_HINT_PROMPT_IDS_SQL)
+        ]
+        prompt_steps = [
+            step for step in plan if " p " in f" {step} " or "TABLE p" in step
+        ]
+        self.assertTrue(prompt_steps, plan)
+        self.assertEqual([step for step in plan if step.startswith("SCAN p")], [])
+        self.assertEqual([step for step in plan if step.startswith("SCAN q")], [])
+
     def test_hints_require_exact_unique_raw_text_and_never_refresh(self):
         db = self.database()
         conn = db.model.get_connection()

tests/test_execution_history_hooks.py

@@ -19,6 +19,7 @@ from unittest.mock import patch
 from tests.history_waits import drain_writer
 from database import PromptDatabase
 from utils import usage_tracking
+from utils.hashing import generate_prompt_hash
 from utils.execution_history_writer import HistoryWriter
 
 try:
@@ -918,6 +919,30 @@ class CachedRuntimeHistoryTests(HooksCase):
         self.assertIsNone(named_prompt(self.db, self.rows()[1]["run_id"]))
 
 
+class CachedRuntimeCaseHistoryTests(CachedRuntimeHistoryTests):
+    """Raw text that differs only in case joins the saved prompt's history (#460).
+
+    The hash folds case, so the run counts toward the lower-case prompt saved
+    first; the run keeps its exact raw spelling.
+    """
+
+    TEXT = "\u00c9 LIGHTHOUSE the LLM wrote"
+
+    def setUp(self):
+        super().setUp()
+        self.db.delete_prompt(self.prompt_id)
+        lower = self.TEXT.lower()
+        self.prompt_id = self.db.save_prompt(
+            text=lower, prompt_hash=generate_prompt_hash(lower)
+        )
+
+    def test_the_library_counts_the_case_variant_as_this_prompt(self):
+        self.assertEqual(
+            self.db.get_prompt_by_hash(generate_prompt_hash(self.TEXT))["id"],
+            self.prompt_id,
+        )
+
+
 class CachedRuntimeWhitespaceHistoryTests(CachedRuntimeHistoryTests):
     """Raw text with outer whitespace joins the history of the prompt it saves (#438).
 

tests/test_foreign_save_to_library.py

@@ -0,0 +1,330 @@
+"""#247 PR 2 x #90: the monitor's foreign path never stores switched-off text.
+
+A file whose run's prompt has Save to library off is the opted-out path's,
+not the foreign path's; a foreign graph whose ladder decision is
+``not_saved`` (a switched-off node the #90 walk does not reach) stores
+nothing at settle, at the re-check after the seal, or from recorded history,
+and leaves no wait behind.
+"""
+
+import os
+import unittest
+
+from tests import saver_anchor_graphs as shapes
+from tests import test_run_attribution as attribution
+from tests import test_save_to_library as save_tests
+from tests.test_pair_scan import sampled_graph
+from utils.foreign_runs import ForeignRecheck
+from utils.hashing import generate_prompt_hash
+from py.config import GalleryConfig  # after tests: it stubs the host
+
+LOGGER = "prompt_manager.image_monitor"
+OFF_TEXT = "SECRET OFF TEXT"
+SKIPPED = "Save to library is off for this run's prompt, skipping image"
+
+# The class dict, not the class: a TestCase bound here would run again
+Fixture = vars(attribution.TestMonitorForeignRuns)
+
+
+def off_behind(middle="MyNode", key="value"):
+    """A switched-off PromptManagerText reaching a core CLIPTextEncode through
+    ``middle``'s ``key`` input: the #90 walk does not follow ``key``."""
+    graph = save_tests.run_graph()
+    graph["1"] = {
+        "class_type": "PromptManagerText",
+        "inputs": {"text": OFF_TEXT, "save_to_library": False},
+    }
+    graph["7"] = {"class_type": middle, "inputs": {key: ["1", 0]}}
+    graph["8"] = {
+        "class_type": "CLIPTextEncode",
+        "inputs": {"text": ["7", 0], "clip": ["4", 1]},
+    }
+    graph["3"]["inputs"]["positive"] = ["8", 0]
+    return graph
+
+
+def off_text_unwired():
+    """Only history can decide; the legacy answer is a dangling switched-off
+    node's own text (the ladder says ``not_saved`` once it may ask history)."""
+    graph = sampled_graph("unused", "blurry")
+    graph["1"]["inputs"]["text"] = ["30", 0]
+    graph["30"] = {"class_type": "VendorPromptMaker", "inputs": {"prompt": OFF_TEXT}}
+    graph["40"] = {
+        "class_type": "PromptManagerText",
+        "inputs": {"text": OFF_TEXT, "save_to_library": False},
+    }
+    return graph
+
+
+class TestForeignPathKeepsSaveToLibrary(attribution.RecordedRuns, unittest.TestCase):
+    setUp = Fixture["setUp"]
+    close_databases = Fixture["close_databases"]
+    wait_for = Fixture["wait_for"]
+    image = Fixture["image"]
+    linked_text = Fixture["linked_text"]
+    key = Fixture["key"]
+    prompt_count = Fixture["prompt_count"]
+    settled = Fixture["settled"]
+
+    def reference(self, path):
+        return {
+            "folder_type": "output",
+            "subfolder": "dataset",
+            "filename": os.path.basename(path),
+        }
+
+    def seal(self, path, positives=(), nodes=("9",)):
+        """Record one run reporting ``path`` (a new run id for every call)."""
+        self.runs = getattr(self, "runs", -1) + 1
+        self.record_run(
+            self.db,
+            self.output,
+            [("dataset", os.path.basename(path))],
+            list(positives),
+            item=self.runs,
+            nodes=list(nodes),
+        )
+
+    def recheck(self, path):
+        """The re-check the seal schedules, run through the monitor's entry."""
+        with self.assertLogs(LOGGER, "INFO") as logs:
+            self.handler.process_new_image(path, ForeignRecheck(self.reference(path)))
+        return "\n".join(logs.output)
+
+    def assert_nothing_stored(self, path):
+        self.assertIsNone(self.linked_text(path))
+        self.assertEqual(self.prompt_count(), 0)
+        self.assertIsNone(self.db.get_prompt_by_hash(generate_prompt_hash(OFF_TEXT)))
+        self.assertEqual(self.handler.awaiting_foreign(), [])
+        self.assertEqual(list(self.handler._foreign_due), [])
+
+    def test_q1_the_only_prompt_manager_node_off_is_opted_out_not_foreign(self):
+        graph = save_tests.run_graph(save=False)
+        # The switched-off positive is the only PromptManager node
+        graph["2"] = {"class_type": "CLIPTextEncode", "inputs": {"text": "blurry"}}
+        sampler_less = {
+            "1": {
+                "class_type": "PromptManagerText",
+                "inputs": {"text": OFF_TEXT, "save_to_library": False},
+            }
+        }
+        for name, graph in (("sampled", graph), ("sampler-less", sampler_less)):
+            with self.subTest(name=name):
+                self.metadata = {"prompt": graph}
+                state, _ = self.handler._find_prompt_from_metadata(self.metadata)
+                self.assertEqual(state, "opted_out")
+                path = self.image(f"IMG_GEN_{name}_00001_.png")
+                with self.assertLogs(LOGGER, "INFO") as logs:
+                    self.handler.process_new_image(path)
+                self.assertIn(SKIPPED, "\n".join(logs.output))
+                self.seal(path)
+                self.assertTrue(self.settled())
+                self.assertIn(SKIPPED, self.recheck(path))
+                self.assert_nothing_stored(path)
+
+    def test_q2_an_off_text_node_behind_a_pass_through_stores_nothing(self):
+        for middle, key in (("MyNode", "value"), ("easy showAnything", "anything")):
+            with self.subTest(middle=middle):
+                self.metadata = {"prompt": off_behind(middle, key)}
+                state, _ = self.handler._find_prompt_from_metadata(self.metadata)
+                self.assertEqual(state, "foreign")
+                path = self.image(f"IMG_GEN_{key}_00001_.png")
+                with self.assertLogs(LOGGER, "INFO") as logs:
+                    self.handler.process_new_image(path)
+                self.assertIn(SKIPPED, "\n".join(logs.output))
+                self.assert_nothing_stored(path)
+                self.seal(path)
+                self.assertTrue(self.settled())
+                self.assertIn(SKIPPED, self.recheck(path))
+                self.assert_nothing_stored(path)
+
+    def test_q2_a_wired_switch_behind_a_pass_through_stores_nothing(self):
+        graph = off_behind()
+        graph["1"]["inputs"]["save_to_library"] = ["20", 0]
+        graph["20"] = {"class_type": "SomeBoolNode", "inputs": {}}
+        self.metadata = {"prompt": graph}
+        path = self.image("IMG_GEN_wired_00001_.png")
+        wired = "Save to library is wired from a node the image cannot show"
+        with self.assertLogs(LOGGER, "INFO") as logs:
+            self.handler.process_new_image(path)
+        self.assertIn(wired, "\n".join(logs.output))
+        self.seal(path)
+        self.assertTrue(self.settled())
+        self.assertIn(wired, self.recheck(path))
+        self.assert_nothing_stored(path)
+
+    def test_q3_an_undecided_not_saved_waits_for_the_record_then_skips(self):
+        """No record yet: a not_saved of the undecided step is not final (a
+        recorded saver may resolve another branch), so the file waits for its
+        run; the decision with the record stores nothing and logs the skip."""
+        wired = off_text_unwired()
+        wired["40"]["inputs"]["save_to_library"] = ["20", 0]
+        wired["20"] = {"class_type": "SomeBoolNode", "inputs": {}}
+        cases = (
+            ("off", off_text_unwired(), SKIPPED),
+            ("wired", wired, "Save to library is wired from a node"),
+        )
+        for name, graph, line in cases:
+            with self.subTest(name=name):
+                self.metadata = {"prompt": graph}
+                path = self.image(f"vendor_{name}_00001_.png")
+                self.handler.process_new_image(path)
+                self.assertIsNone(self.linked_text(path))
+                self.assertEqual(self.handler.awaiting_foreign(), [self.key(path)])
+                with self.assertLogs(LOGGER, "INFO") as logs:
+                    self.seal(path)
+                    self.assertTrue(
+                        self.wait_for(lambda: not self.handler.awaiting_foreign())
+                    )
+                    self.assertTrue(self.settled())
+                self.assertIn(line, "\n".join(logs.output))
+                self.assert_nothing_stored(path)
+
+    def test_q3_an_early_not_saved_never_waits_even_with_a_saver_branch(self):
+        """Review: a switched-off node feeding one branch's positive is the
+        ladder's early, final not_saved (PR 1's #90 rule covers every
+        sampler's positive), so the file never waits and nothing is stored,
+        not even the plain branch the recorded saver names."""
+        graph = shapes.shared_prefix()
+        graph["20"]["inputs"]["text"] = ["50", 0]
+        graph["50"] = {
+            "class_type": "PromptManagerText",
+            "inputs": {"text": OFF_TEXT, "save_to_library": False},
+        }
+        self.metadata = {"prompt": graph}
+        path = self.image("IMG_GEN_00001_.png")
+        with self.assertLogs(LOGGER, "INFO") as logs:
+            self.handler.process_new_image(path)
+        self.assertIn(SKIPPED, "\n".join(logs.output))
+        self.assert_nothing_stored(path)
+        self.seal(path, nodes=("13",))
+        self.assertTrue(self.settled())
+        self.assertIn(SKIPPED, self.recheck(path))
+        self.assert_nothing_stored(path)
+        self.assertIsNone(
+            self.db.get_prompt_by_hash(generate_prompt_hash(shapes.BRANCH_TEXTS[0]))
+        )
+
+    def test_q3_an_off_node_in_the_recorded_saver_branch_is_never_foreign(self):
+        """PR 3's parity case: a switched-off node feeds the very branch the
+        recorded saver names (saver "23" is branch 2). A literal switch is the
+        opted-out path's; a wired one is the node's own run's (#90), never the
+        foreign path's. Nothing is stored at settle or after the run, and no
+        file waits for a foreign re-check."""
+        cases = (
+            ("off", False, "opted_out", SKIPPED),
+            ("wired", ["52", 0], "unknown", "No certain prompt for"),
+        )
+        for name, save, state, line in cases:
+            with self.subTest(name=name):
+                graph = shapes.shared_prefix()
+                graph["50"] = {
+                    "class_type": "PromptManagerText",
+                    "inputs": {"text": OFF_TEXT, "save_to_library": save},
+                }
+                graph["20"]["inputs"]["text"] = ["50", 0]
+                if isinstance(save, list):
+                    graph["52"] = {"class_type": "SomeBoolNode", "inputs": {}}
+                self.metadata = {"prompt": graph}
+                found, _ = self.handler._find_prompt_from_metadata(self.metadata)
+                self.assertEqual(found, state)
+                path = self.image(f"IMG_GEN_{name}_00001_.png")
+                with self.assertLogs(LOGGER, "INFO") as logs:
+                    self.handler.process_new_image(path)
+                self.assertIn(line, "\n".join(logs.output))
+                self.assert_nothing_stored(path)
+                self.seal(path, positives=[OFF_TEXT], nodes=("23",))
+                self.assertTrue(self.settled())
+                self.handler.process_new_image(path)
+                self.assert_nothing_stored(path)
+                self.assertIsNone(
+                    self.db.get_prompt_by_hash(
+                        generate_prompt_hash(shapes.BRANCH_TEXTS[1])
+                    )
+                )
+
+    def test_q4_an_off_node_on_the_negative_only_still_links_the_positive(self):
+        positive = "a lighthouse on a cliff at dusk"
+        plain = sampled_graph(positive, "unused")
+        plain["2"] = {
+            "class_type": "PromptManager",
+            "inputs": {"text": OFF_TEXT, "save_to_library": False},
+        }
+        behind = sampled_graph(positive, "unused")
+        behind["2"]["inputs"]["text"] = ["7", 0]
+        behind["7"] = {
+            "class_type": "PromptManagerText",
+            "inputs": {"text": OFF_TEXT, "save_to_library": False},
+        }
+        for name, graph in (("direct", plain), ("behind an encoder", behind)):
+            with self.subTest(name=name):
+                self.metadata = {"prompt": graph}
+                state, _ = self.handler._find_prompt_from_metadata(self.metadata)
+                self.assertEqual(state, "foreign")
+                path = self.image(f"plain_{len(name)}_00001_.png")
+                self.handler.process_new_image(path)
+                self.assertEqual(self.linked_text(path), positive)
+                self.assertIsNone(
+                    self.db.get_prompt_by_hash(generate_prompt_hash(OFF_TEXT))
+                )
+                pairs = self.db.list_prompt_pairs()["pairs"]
+                self.assertEqual(pairs, [])
+                self.assertEqual(self.handler.awaiting_foreign(), [])
+
+    def test_q5_recorded_history_never_brings_back_switched_off_text(self):
+        """Even a recorded run whose contexts hold the off text (an older
+        history row) cannot name it, at settle or at the re-check."""
+        cases = {"pass-through": off_behind(), "dangling": off_text_unwired()}
+        for name, graph in cases.items():
+            with self.subTest(name=name):
+                self.metadata = {"prompt": graph}
+                path = self.image(f"IMG_GEN_{len(name)}_00001_.png")
+                self.seal(path, positives=[OFF_TEXT], nodes=("9",))
+                self.handler.process_new_image(path)
+                self.assert_nothing_stored(path)
+                self.assertIn(SKIPPED, self.recheck(path))
+                self.assert_nothing_stored(path)
+
+    def test_q6_with_foreign_saving_off_save_to_library_is_unchanged(self):
+        GalleryConfig.SAVE_FOREIGN_RUNS = False
+        opted_out = save_tests.run_graph(save=False)
+        self.metadata = {"prompt": opted_out}
+        skipped = self.image("IMG_GEN_off_00001_.png")
+        with self.assertLogs(LOGGER, "INFO") as logs:
+            self.handler.process_new_image(skipped)
+        self.assertIn(SKIPPED, "\n".join(logs.output))
+        self.metadata = {"prompt": off_behind()}
+        foreign = self.image("IMG_GEN_foreign_00001_.png")
+        with self.assertLogs(LOGGER, "INFO") as logs:
+            self.handler.process_new_image(foreign)
+        self.assertIn("No PromptManager prompt in workflow", "\n".join(logs.output))
+        for path in (skipped, foreign):
+            self.assert_nothing_stored(path)
+        # A switched-on PromptManager positive still links by its metadata
+        stored = self.db.save_prompt(
+            text=save_tests.POSITIVE,
+            prompt_hash=generate_prompt_hash(save_tests.POSITIVE),
+        )
+        self.metadata = {"prompt": save_tests.run_graph(save=True)}
+        on = self.image("IMG_GEN_on_00001_.png")
+        self.handler.process_new_image(on)
+        self.assertEqual(self.linked_text(on), save_tests.POSITIVE)
+        self.assertEqual(
+            self.db.get_prompt_by_hash(generate_prompt_hash(save_tests.POSITIVE))["id"],
+            stored,
+        )
+
+    def test_the_shared_prefix_dataset_still_links_after_the_seal(self):
+        """Regression guard (passes before the fix): the wait still works."""
+        self.metadata = {"prompt": shapes.shared_prefix()}
+        path = self.image("IMG_GEN_00001_.png")
+        self.handler.process_new_image(path)
+        self.assertEqual(self.handler.awaiting_foreign(), [self.key(path)])
+        self.seal(path, nodes=("13",))
+        self.assertTrue(self.wait_for(lambda: self.linked_text(path)))
+        self.assertTrue(self.settled())
+        self.assertEqual(self.linked_text(path), shapes.BRANCH_TEXTS[0])
+
+
+if __name__ == "__main__":
+    unittest.main()

tests/test_hashing.py

@@ -17,6 +17,8 @@ from utils.hashing import (
     generate_prompt_hash,
     is_duplicate_prompt,
     STRIPPED_CODE_POINTS,
+    prompt_key,
+    register_prompt_functions,
     stored_prompt_text,
     stored_prompt_text_sql,
 )
@@ -158,5 +160,56 @@ class TestStoredPromptText(unittest.TestCase):
                 self.assertEqual(self.sql_strip(value), stored_prompt_text(value))
 
 
+class TestPromptKey(unittest.TestCase):
+    """History compares the hash's own key; SQL calls Python for it (#460)."""
+
+    CASES = (
+        "A Lighthouse",
+        " \u00c9t\u00c9\n",
+        "\u03a3\u03a3",
+        "\u0130stanbul",
+        "\u1e9e",
+        "\u212a",
+        "ok",
+    )
+
+    def test_key_is_the_hash_normalization(self):
+        import hashlib
+
+        for value in self.CASES:
+            with self.subTest(value=repr(value)):
+                self.assertEqual(
+                    generate_prompt_hash(value),
+                    hashlib.sha256(prompt_key(value).encode("utf-8")).hexdigest(),
+                )
+        self.assertEqual(prompt_key("  A Lighthouse\n"), "a lighthouse")
+
+    def test_sql_functions_give_pythons_result_where_sql_lower_does_not(self):
+        import sqlite3
+
+        conn = sqlite3.connect(":memory:")
+        try:
+            register_prompt_functions(conn)
+            for value in self.CASES:
+                with self.subTest(value=repr(value)):
+                    key, digest = conn.execute(
+                        "SELECT pm_prompt_key(?), pm_prompt_hash(?)", (value, value)
+                    ).fetchone()
+                    self.assertEqual(key, prompt_key(value))
+                    self.assertEqual(digest, generate_prompt_hash(value))
+            self.assertEqual(
+                conn.execute(
+                    "SELECT pm_prompt_key(1), pm_prompt_hash(NULL)"
+                ).fetchone(),
+                (None, None),
+            )
+            # SQLite's own lower() folds ASCII only
+            self.assertNotEqual(
+                conn.execute("SELECT lower(?)", ("\u00c9",)).fetchone()[0], "\u00e9"
+            )
+        finally:
+            conn.close()
+
+
 if __name__ == "__main__":
     unittest.main()

tests/test_history_reads.py

@@ -17,6 +17,7 @@ from utils.execution_contexts import (
 )
 from utils.execution_outputs import OutputBundle
 from tests.test_execution_context_database import bundle, snapshot
+from utils.hashing import generate_prompt_hash
 from tests.test_execution_output_database import bundle as outputs
 from tests import test_execution_history_database as history_tests
 
@@ -33,6 +34,10 @@ class HistoryReadTests(unittest.TestCase):
             conn.execute("INSERT INTO prompts(id,text) VALUES(?,?)", (identity, text))
         return identity
 
+    def hashed(self, db, text):
+        """A prompt saved as the nodes save it: with its dedup hash."""
+        return db.save_prompt(text, prompt_hash=generate_prompt_hash(text))
+
     def capture(self, db, item=0, context=None, output=None, time=100):
         start = self.start(db, item, time)
         db.write_execution_start(start)
@@ -151,9 +156,9 @@ class HistoryReadTests(unittest.TestCase):
             )
         self.require_reads(db)
         with db.recorded_history_snapshot():
-            # Outer whitespace joins the stored form's scope (#438); case and
-            # Unicode normalization stay distinct.
-            expected = (4, 1, 4, 4, 1, 1, 2, 2)
+            # Outer whitespace (#438) and case (#460) join one scope;
+            # Unicode normalization stays distinct.
+            expected = (5, 5, 5, 5, 1, 1, 2, 2)
             for i, count in enumerate(expected, 1):
                 self.assertEqual(db.list_history_window(i)["recorded_run_count"], count)
             self.assertIsNone(db.list_history_window(100))
@@ -1047,3 +1052,70 @@ class HistoryReadTests(unittest.TestCase):
         with db.recorded_history_snapshot():
             found = db.get_newest_pair_run(positive, pair)
         self.assertEqual(found["run"]["run_id"], newest.run_id)
+
+    def test_case_variant_raw_text_joins_the_saved_prompt_scope(self):
+        # #460: the hash folds case, so these runs count toward this prompt.
+        db = self.database()
+        positive = self.hashed(db, "\u00e9t\u00e9 lighthouse \u03c3")
+        raws = (
+            "\u00c9T\u00c9 Lighthouse \u03a3\n",
+            "\u00e9t\u00e9 lighthouse \u03c3",
+            " \u00c9t\u00e9 LIGHTHOUSE \u03a3",
+        )
+        starts = [
+            self.capture(db, i, bundle(positive=raw), time=100 + i)
+            for i, raw in enumerate(raws)
+        ]
+        # Unicode normalization and inner text stay distinct
+        self.capture(db, 7, bundle(positive="e\u0301te\u0301 lighthouse \u03c3"))
+        self.capture(db, 8, bundle(positive="\u00e9t\u00e9  lighthouse \u03c3"))
+        dotted = self.hashed(db, "\u0130stanbul")
+        dotted_run = self.capture(db, 9, bundle(positive="\u0130STANBUL"), time=300)
+        with db.recorded_history_snapshot():
+            page = db.list_history_window(positive)
+            details = [db.get_history_run(start.run_id) for start in starts]
+            turkish = db.list_history_window(dotted)
+        self.assertEqual(page["recorded_run_count"], 3)
+        self.assertEqual(
+            [r["run_id"] for r in page["items"]], [s.run_id for s in starts]
+        )
+        for raw, detail in zip(raws, details):
+            context = detail["contexts"][0]
+            self.assertEqual(context["positive"]["raw"], raw)
+            self.assertEqual(
+                context["positive_raw_text_id"], page["group"]["raw_text_id"]
+            )
+            self.assertEqual(context["positive_prompt_id"], positive)
+        self.assertEqual([r["run_id"] for r in turkish["items"]], [dotted_run.run_id])
+
+    def test_newest_pair_run_follows_case_variant_raw_text(self):
+        db = self.database()
+        pair = db.save_prompt_pair("a lighthouse", "fog")
+        positive = db.get_prompt_pair(pair)["positive_prompt_id"]
+        newest = self.capture(db, 0, bundle("FOG", positive="A Lighthouse"), time=100)
+        with db.recorded_history_snapshot():
+            found = db.get_newest_pair_run(positive, pair)
+        self.assertEqual(found["run"]["run_id"], newest.run_id)
+
+    def test_a_position_from_before_460_resumes_the_same_scope(self):
+        db = self.database()
+        positive = self.hashed(db, "A Lighthouse")
+        for i in range(3):
+            self.capture(db, i, bundle(positive="a lighthouse"), time=100 + i)
+        with db.recorded_history_snapshot():
+            first = db.list_history_window(positive, limit=2)
+            position = first["next_position"]
+            legacy = replace(
+                position, raw_text_id=content_digest("text", ("A Lighthouse",))
+            )
+            rest = db.list_history_window(positive, limit=2, after=legacy)
+            with self.assertRaises(ValueError):
+                db.list_history_window(
+                    positive,
+                    after=replace(position, raw_text_id=content_digest("text", ("x",))),
+                )
+        self.assertEqual(first["recorded_run_count"], 3)
+        self.assertEqual(len(rest["items"]), 1)
+        self.assertEqual(
+            rest["group"]["raw_text_id"], content_digest("text", ("a lighthouse",))
+        )

tests/test_image_linking.py

@@ -172,17 +172,40 @@ class TestForeignWorkflows(LinkingTestCase):
         self.tracker.set_current_prompt(text, {"prompt_id": prompt_id})
         return prompt_id
 
+    def _linked_text(self):
+        row = (
+            self.db.model.get_connection()
+            .execute(
+                "SELECT p.text FROM generated_images g "
+                "JOIN prompts p ON p.id = g.prompt_id WHERE g.image_path = ?",
+                (self.image,),
+            )
+            .fetchone()
+        )
+        return row[0] if row else None
+
     def test_foreign_image_leaves_the_batch_queue_aligned(self):
+        """#247: it links to its own prompt and never takes a queue entry."""
         first = self._queue_batch_item("item one")
         self.assertIsNone(self._process({"prompt": foreign_graph()}))
         self.assertEqual(self.linked, [])
         self.assertEqual(self.tracker.pop_next_prompt()["id"], first)
         self.assertIsNone(self.tracker.pop_next_prompt())
+        self.assertEqual(self._linked_text(), "a cat")
+        prompt = self.db.get_prompt_by_hash(generate_prompt_hash("a cat"))
+        self.assertEqual(prompt["run_count"] or 0, 0)
 
-    def test_foreign_image_is_not_linked_by_the_later_fallbacks_either(self):
+    def test_a_foreign_image_is_linked_only_by_the_ladder(self):
         self._queue_batch_item("item one")  # live tracker and recent prompt exist
+        untraceable = foreign_graph()
+        untraceable["9"]["inputs"]["text"] = ["30", 0]
+        untraceable["30"] = {"class_type": "VendorPromptMaker", "inputs": {}}
+        self.assertIsNone(self._process({"prompt": untraceable}))
+        self.assertEqual(self.linked, [])
+        self.assertIsNone(self._linked_text())
         self.assertIsNone(self._process({"prompt": foreign_graph()}))
         self.assertEqual(self.linked, [])
+        self.assertEqual(self._linked_text(), "a cat")
 
     def test_negative_only_workflow_is_foreign(self):
         self._save("blurry")

tests/test_image_monitor_singleton.py

@@ -597,6 +597,11 @@ class TestHandlerScheduling(unittest.TestCase):
             self.handler._schedule_recheck(path)
         self.assertIsNone(self.handler._worker)
         self.assertEqual(list(self.handler._history_due), [])
+        # #247: a foreign file's sealed re-check is dropped the same way
+        self.handler._await_foreign(path)
+        self.assertEqual(self.handler.awaiting_foreign(), [])
+        self.handler._schedule_recheck(path, {"filename": "late.png"})
+        self.assertEqual(list(self.handler._foreign_due), [])
         self.assertEqual(self._live_workers() - before, set())
 
     def test_a_sealed_run_starts_a_worker_on_an_open_handler(self):

tests/test_inspection_response.py

@@ -887,6 +887,36 @@ class TestImageTagsFacet(unittest.TestCase):
         self.assertNotIn("image_tags", json.loads(encode(inspection({}))))
         self.assertNotIn("image_tags", self.encoded(None))
 
+    def history_encoded(self, facet):
+        from utils.inspection_response import encode_history_inspection_response
+
+        body = encode_history_inspection_response(
+            current_output={"history_epoch": "epoch"},
+            file={"observation": "current-file", "size": 100},
+            prompts=PROMPTS,
+            resolution={"state": "resolved", "sampler_id": "3", "reason": ""},
+            inspection=inspection({"prompt": graph_text()}),
+            library={"status": "none"},
+            image_tags=facet,
+        )
+        return json.loads(body)
+
+    def test_history_envelope_validates_image_tags_like_the_rooted_one(self):
+        """#457: History's current-file read carries the same facet."""
+        for facet in (
+            {"state": "current", "tags": ["dunes", "warm_light"]},
+            {"state": "none", "tags": []},
+            {"state": "unavailable", "tags": []},
+            {"state": "current", "tags": ["a,b"]},
+            {"state": "current", "tags": [], "locator": "/abs/path.png"},
+        ):
+            with self.subTest(facet=facet):
+                self.assertEqual(
+                    self.history_encoded(facet)["image_tags"],
+                    self.encoded(facet)["image_tags"],
+                )
+        self.assertNotIn("image_tags", self.history_encoded(None))
+
 
 class TestVideoProbeBoundsMatchPngBudgets(unittest.TestCase):
     """ffprobe escapes tag text; the stdout bounds must never gate a record

tests/test_library_folders.py

@@ -0,0 +1,243 @@
+"""#444: the Library's output-folder filter and multi-tag OR."""
+
+import ntpath
+import os
+import posixpath
+import tempfile
+import unittest
+from pathlib import Path
+
+from database.library_folders import (
+    folder_prefixes,
+    normalize_folder,
+    relative_folder,
+)
+from database.operations import PromptDatabase
+from utils.hashing import generate_prompt_hash
+
+
+class TestFolderNames(unittest.TestCase):
+    def test_normalize_strips_outer_slashes_on_both_flavors(self):
+        for flavor in (posixpath, ntpath):
+            with self.subTest(flavor=flavor.__name__):
+                self.assertEqual(normalize_folder("", flavor), "")
+                self.assertEqual(normalize_folder("/", flavor), "")
+                self.assertEqual(normalize_folder("/a/b/", flavor), "a/b")
+                self.assertEqual(normalize_folder("..foo/b", flavor), "..foo/b")
+
+    def test_a_backslash_separates_only_on_windows(self):
+        # Live B1: on POSIX a backslash is part of a folder name, as listed.
+        name = "Metadata\\Klein_4b Wildcard Run"
+        self.assertEqual(normalize_folder(name, posixpath), name)
+        self.assertEqual(
+            normalize_folder("2026\\08-Aug\\", posixpath), "2026\\08-Aug\\"
+        )
+        self.assertEqual(
+            normalize_folder(name, ntpath), "Metadata/Klein_4b Wildcard Run"
+        )
+        self.assertEqual(normalize_folder("2026\\08-Aug\\", ntpath), "2026/08-Aug")
+        # A colon names a drive only on Windows.
+        self.assertEqual(normalize_folder("a:b/c", posixpath), "a:b/c")
+
+    def test_normalize_refuses_anything_but_a_folder_below_a_root(self):
+        for flavor, value in (
+            *(
+                (flavor, value)
+                for flavor in (posixpath, ntpath)
+                for value in ("a/../b", "./a", "a//b", "a\x00b", 3, "x" * 1025)
+            ),
+            (ntpath, "C:/x"),
+            (ntpath, "c:"),
+            (ntpath, "a\\..\\b"),
+        ):
+            with self.subTest(flavor=flavor.__name__, value=value):
+                with self.assertRaises(ValueError):
+                    normalize_folder(value, flavor)
+
+    def test_the_default_flavor_is_the_platform_path_module(self):
+        self.assertEqual(normalize_folder("a\\b"), normalize_folder("a\\b", os.path))
+
+    def test_posix_prefixes_are_exact_and_end_in_a_separator(self):
+        self.assertEqual(
+            folder_prefixes(["/out", "/data/out/"], "a/b", posixpath),
+            ["/out/a/b/", "/data/out/a/b/"],
+        )
+        self.assertEqual(folder_prefixes(["/"], "a", posixpath), ["/a/"])
+        self.assertEqual(
+            folder_prefixes(["/out", "/out/"], "a", posixpath), ["/out/a/"]
+        )
+        with self.assertRaises(ValueError):
+            folder_prefixes([], "a", posixpath)
+
+    def test_windows_prefixes_fold_case_including_non_ascii(self):
+        self.assertEqual(
+            folder_prefixes(["C:\\ComfyUI\\Output"], "Ä/Sub", ntpath),
+            ["c:\\comfyui\\output\\ä\\sub\\"],
+        )
+        self.assertEqual(folder_prefixes(["D:\\"], "x", ntpath), ["d:\\x\\"])
+        self.assertEqual(
+            folder_prefixes(["\\\\Server\\Share\\Out"], "a", ntpath),
+            ["\\\\server\\share\\out\\a\\"],
+        )
+        # The stored file_path is normcase(normpath(abspath)), the same fold.
+        stored = ntpath.normcase(ntpath.normpath("C:/ComfyUI/output/Ä/sub/x.png"))
+        self.assertTrue(
+            stored.startswith(
+                folder_prefixes(["C:/ComfyUI/Output"], "ä/SUB", ntpath)[0]
+            )
+        )
+
+    def test_relative_folder_keeps_the_stored_spelling(self):
+        self.assertEqual(
+            relative_folder("/out/2026/Aug/x.png", ["/out"], posixpath), "2026/Aug"
+        )
+        self.assertIsNone(relative_folder("/out/x.png", ["/out"], posixpath))
+        self.assertIsNone(relative_folder("/outside/x.png", ["/out"], posixpath))
+        self.assertIsNone(relative_folder("/output2/a/x.png", ["/out"], posixpath))
+        self.assertEqual(
+            relative_folder("/out/..foo/x.png", ["/out"], posixpath), "..foo"
+        )
+        self.assertEqual(
+            relative_folder("C:\\Out\\Ä\\Sub\\x.png", ["c:\\out"], ntpath), "Ä/Sub"
+        )
+        self.assertEqual(relative_folder("D:\\a\\x.png", ["D:\\"], ntpath), "a")
+
+
+class TestLibraryFolderAndAnyTag(unittest.TestCase):
+    def setUp(self):
+        self.temp = tempfile.TemporaryDirectory()
+        base = Path(os.path.realpath(self.temp.name))
+        self.root = base / "output"
+        self.root.mkdir()
+        self.db = PromptDatabase(str(base / "folders.db"))
+
+    def tearDown(self):
+        try:
+            self.db.close_all()
+        finally:
+            self.temp.cleanup()
+
+    def save(self, text, **metadata):
+        return self.db.save_prompt(
+            text, prompt_hash=generate_prompt_hash(text), **metadata
+        )
+
+    def link(self, prompt_id, *parts):
+        path = self.root.joinpath(*parts)
+        path.parent.mkdir(parents=True, exist_ok=True)
+        path.write_bytes(b"x")
+        self.db.link_image_to_prompt(prompt_id, str(path))
+
+    def texts(self, **options):
+        return sorted(item["text"] for item in self.db.list_library(**options)["items"])
+
+    def in_folder(self, folder, roots=None):
+        return self.texts(
+            folder_prefixes=folder_prefixes(roots or [str(self.root)], folder)
+        )
+
+    def test_folder_matches_its_descendants_only(self):
+        self.link(self.save("deep"), "a", "b", "x.png")
+        self.link(self.save("top"), "a", "y.png")
+        self.link(self.save("sibling"), "ab", "z.png")
+        self.link(self.save("literal"), "100%_off", "w.png")
+        self.save("no images")
+        self.assertEqual(self.in_folder("a"), ["deep", "top"])
+        self.assertEqual(self.in_folder("a/b"), ["deep"])
+        self.assertEqual(self.in_folder("ab"), ["sibling"])
+        self.assertEqual(self.in_folder("100%_off"), ["literal"])
+        self.assertEqual(self.in_folder("1"), [])
+        cut = self.db.list_library(
+            folder_prefixes=folder_prefixes([str(self.root)], "a"), limit=1
+        )
+        self.assertEqual(cut["total"], 2)
+
+    def test_folder_combines_with_text_category_and_tags(self):
+        self.link(self.save("red cat", category="Art", tags=["cat"]), "a", "1.png")
+        self.link(self.save("red dog", category="Art", tags=["dog"]), "a", "2.png")
+        self.link(self.save("red bird", category="Art", tags=["cat"]), "b", "3.png")
+        prefixes = folder_prefixes([str(self.root)], "a")
+        self.assertEqual(
+            self.texts(folder_prefixes=prefixes, tags=["cat"]), ["red cat"]
+        )
+        self.assertEqual(
+            self.texts(folder_prefixes=prefixes, text="dog", category="Art"),
+            ["red dog"],
+        )
+
+    @unittest.skipIf(os.name == "nt", "symlinks need privileges on Windows")
+    def test_folder_listing_and_match_through_a_symlinked_root(self):
+        link = self.root.parent / "linked"
+        os.symlink(self.root, link)
+        prompt = self.save("via link")
+        (self.root / "a").mkdir()
+        (self.root / "a" / "x.png").write_bytes(b"x")
+        self.db.link_image_to_prompt(prompt, str(link / "a" / "x.png"))
+        spellings = [str(self.root), str(link)]
+        self.assertEqual(self.in_folder("a", spellings), ["via link"])
+        self.assertEqual(self.in_folder("a", [str(self.root)]), [])
+        self.assertEqual(self.db.get_library_options(spellings)["folders"], ["a"])
+
+    @unittest.skipIf(os.name == "nt", "a backslash is a separator on Windows")
+    def test_a_posix_folder_with_a_backslash_lists_and_filters_alike(self):
+        name = "Metadata\\Klein_4b Wildcard Run"
+        self.link(self.save("klein"), name, "x.png")
+        self.link(self.save("other"), "Metadata", "y.png")
+        listed = self.db.get_library_options([str(self.root)])["folders"]
+        self.assertEqual(listed, ["Metadata", name])
+        self.assertEqual(self.in_folder(normalize_folder(name)), ["klein"])
+        self.assertEqual(self.in_folder(normalize_folder("Metadata")), ["other"])
+
+    def test_folder_listing_offers_ancestors_and_skips_outside_images(self):
+        self.link(self.save("deep"), "2026", "08-Aug", "x.png")
+        self.link(self.save("root image"), "y.png")
+        outside = Path(self.temp.name) / "elsewhere" / "z.png"
+        outside.parent.mkdir()
+        outside.write_bytes(b"x")
+        self.db.link_image_to_prompt(self.save("outside"), str(outside))
+        options = self.db.get_library_options([str(self.root)])
+        self.assertEqual(options["folders"], ["2026", "2026/08-Aug"])
+        self.assertNotIn("folders", self.db.get_library_options())
+
+    def test_folder_combines_with_untagged(self):
+        self.link(self.save("plain in a"), "a", "1.png")
+        self.link(self.save("tagged in a", tags=["t"]), "a", "2.png")
+        self.link(self.save("plain in b"), "b", "3.png")
+        prefixes = folder_prefixes([str(self.root)], "a")
+        self.assertEqual(
+            self.texts(folder_prefixes=prefixes, tag_mode="untagged"), ["plain in a"]
+        )
+
+    def test_any_tag_mode_is_the_union_across_parent_and_pair(self):
+        self.save("parent a", tags=["a"])
+        self.save("parent b", tags=["b"])
+        self.save("both", tags=["a", "b"])
+        self.save("neither", tags=["c"])
+        self.db.save_prompt_pair("pair b", "noise", tags=["b"])
+        split = self.save("split", tags=["a"])
+        self.db.save_prompt_pair("split", "split noise", tags=["b"])
+        archived = self.db.save_prompt_pair("archived b", "old", tags=["b"])
+        self.db.archive_prompt_pair(archived)
+        self.assertIsInstance(split, int)
+        self.assertEqual(
+            self.texts(tags=["a", "b"], tag_mode="any"),
+            ["both", "pair b", "parent a", "parent b", "split"],
+        )
+        self.assertEqual(self.texts(tags=["a", "b"], tag_mode="all"), ["both"])
+        self.assertEqual(self.texts(tags=["a", "b"]), ["both"])
+        for tags in (["a"], ["b"], ["c"]):
+            self.assertEqual(
+                self.texts(tags=tags, tag_mode="any"), self.texts(tags=tags)
+            )
+        self.assertEqual(
+            self.texts(tags=["a", "b"], tag_mode="any", text="parent"),
+            ["parent a", "parent b"],
+        )
+        self.assertEqual(self.texts(tag_mode="any"), self.texts())
+        for mode in ("none", None, "ANY"):
+            with self.subTest(mode=mode), self.assertRaises(ValueError):
+                self.db.list_library(tag_mode=mode)
+
+
+if __name__ == "__main__":
+    unittest.main()

tests/test_repair_foreign_links.py

@@ -0,0 +1,951 @@
+"""tools/repair_foreign_links.py: the reviewed repair of the #247 dataset links.
+
+The fixture is a library whose foreign (no PromptManager node) dataset run
+wrote shared-prefix files that the old scan linked to the wrong branch text.
+Recorded history names each file's saver, so the ladder (``recorded-run``)
+names the right branch.
+"""
+
+import contextlib
+import hashlib
+import io
+import json
+import os
+import shutil
+import sqlite3
+import tempfile
+import unittest
+import unittest.mock
+from pathlib import Path
+
+from PIL import Image
+from PIL.PngImagePlugin import PngInfo
+
+try:
+    from tests.open_handles import assert_closed
+except ImportError:  # discovered with tests/ as the top-level directory
+    from open_handles import assert_closed
+
+from database.models import PromptModel
+from database.operations import PromptDatabase
+from tests import saver_anchor_graphs as shapes
+from tests.test_run_attribution import RecordedRuns
+from tools import repair_foreign_links as tool
+from utils.file_prompt import Decision
+from utils.hashing import generate_prompt_hash
+
+TEXTS = shapes.BRANCH_TEXTS
+NAME = "IMG_GEN_{:05}_.png"
+
+
+def png(path, graph):
+    info = PngInfo()
+    info.add_text("prompt", json.dumps(graph))
+    Image.new("RGB", (4, 4)).save(path, pnginfo=info)
+
+
+def sha(text):
+    return hashlib.sha256(text.encode("utf-8")).hexdigest()
+
+
+def file_sha(path):
+    return hashlib.sha256(Path(path).read_bytes()).hexdigest()
+
+
+def dump(path):
+    """The database's logical content (independent of WAL checkpoints)."""
+    conn = sqlite3.connect(path)
+    try:
+        return list(conn.iterdump())
+    finally:
+        conn.close()
+
+
+def not_running(_comfyui_dir, _port):
+    return []
+
+
+def no_handles(_database):
+    return []
+
+
+class RepairFixture(RecordedRuns, unittest.TestCase):
+    """Rows: A relink to a new prompt, B relink to a stored prompt (with a
+    pair link), C keep, D file gone, E two savers (skip), F the UNIQUE case
+    (a wrong row beside a correct row), H relink from a rated prompt, G a
+    PromptManager run (not a candidate)."""
+
+    def setUp(self):
+        self.tmp = tempfile.mkdtemp()
+        self.addCleanup(shutil.rmtree, self.tmp, True)
+        self.addCleanup(assert_closed, self, self.tmp)
+        self.addCleanup(PromptModel.reset_schema_cache)
+        self.output = os.path.join(self.tmp, "output")
+        os.makedirs(os.path.join(self.output, "dataset"))
+        self.db_path = os.path.join(self.tmp, "prompts.db")
+        self.plan_path = os.path.join(self.tmp, "plan.json")
+        self.backups = os.path.join(self.tmp, "backups")
+        os.makedirs(self.backups)
+        self.comfy = os.path.join(self.tmp, "comfy")
+        os.makedirs(self.comfy)
+        Path(self.comfy, "main.py").write_text("# ComfyUI\n", encoding="utf-8")
+        self.files = {}
+        for key, number in zip("ABCDEFHG", range(1, 9)):
+            path = os.path.join(self.output, "dataset", NAME.format(number))
+            png(path, shapes.shared_prefix())
+            self.files[key] = path
+        db = PromptDatabase(self.db_path)
+        try:
+            self.build(db)
+        finally:
+            db.close_all()
+        os.remove(self.files["D"])
+
+    def build(self, db):
+        save = lambda text, **kw: db.save_prompt(  # noqa: E731
+            text=text, prompt_hash=generate_prompt_hash(text), **kw
+        )
+        self.p0 = save(TEXTS[0], category="scanned", tags=["auto-scanned"])
+        self.p2 = save(TEXTS[2], category="scanned", tags=["auto-scanned"])
+        self.orphan = save(
+            "an old guess",
+            category="scanned",
+            tags=["auto-scanned"],
+            notes="Auto-scanned from IMG_GEN_00001_.png",
+        )
+        self.rated = save("a rated guess", rating=5)
+        link = lambda prompt, key: db.link_image_to_prompt(  # noqa: E731
+            prompt, self.files[key], metadata={}
+        )
+        self.rows = {
+            "A": link(self.orphan, "A"),
+            "B": link(self.p0, "B"),
+            "C": link(self.p0, "C"),
+            "D": link(self.p0, "D"),
+            "E": link(self.p0, "E"),
+            "F": link(self.p0, "F"),
+            "F_ok": link(self.p2, "F"),
+            "H": link(self.rated, "H"),
+            "G": link(self.p0, "G"),
+        }
+        conn = db.model.get_connection()
+        with conn:
+            pair = conn.execute(
+                "INSERT INTO prompt_pairs(positive_prompt_id) VALUES(?)", (self.p0,)
+            ).lastrowid
+            conn.execute(
+                "INSERT INTO image_prompt_pairs(image_id, pair_id, provenance) "
+                "VALUES(?,?,'{}')",
+                (self.rows["B"], pair),
+            )
+        name = lambda key: ("dataset", os.path.basename(self.files[key]))  # noqa
+        # Savers of the shared-prefix graph: "13", "23", "33", "43" (branches 1-4)
+        self.record_run(
+            db,
+            self.output,
+            [name(k) for k in "CDEAEBFH"],
+            [],
+            nodes=["13", "13", "13", "23", "23", "33", "33", "43"],
+        )
+        self.record_run(db, self.output, [name("G")], [TEXTS[0]], item=1, nodes=["23"])
+
+    def dry_run(self, *args):
+        out = io.StringIO()
+        code = tool.main(
+            [
+                self.db_path,
+                "--output-dir",
+                self.output,
+                "--plan",
+                self.plan_path,
+                *args,
+            ],
+            out=out,
+        )
+        return code, out.getvalue()
+
+    def apply(self, *args, probe=not_running, handles=no_handles):
+        out = io.StringIO()
+        code = tool.main(
+            [
+                self.db_path,
+                "--apply",
+                self.plan_path,
+                "--comfyui-dir",
+                os.path.join(self.tmp, "comfy"),
+                "--backup-dir",
+                self.backups,
+                *args,
+            ],
+            out=out,
+            probe=probe,
+            handles=handles,
+        )
+        return code, out.getvalue()
+
+    def plan(self):
+        with open(self.plan_path, encoding="utf-8") as stream:
+            return json.load(stream)
+
+    def write_plan(self, plan):
+        with open(self.plan_path, "w", encoding="utf-8") as stream:
+            json.dump(plan, stream)
+
+    def query(self, sql, args=()):
+        conn = sqlite3.connect(self.db_path)
+        try:
+            return conn.execute(sql, args).fetchall()
+        finally:
+            conn.close()
+
+    def query_write(self, sql, args):
+        conn = sqlite3.connect(self.db_path)
+        try:
+            with conn:
+                conn.execute(sql, args)
+        finally:
+            conn.close()
+
+    def links(self):
+        return dict(self.query("SELECT id, prompt_id FROM generated_images"))
+
+
+class DryRunTests(RepairFixture):
+    def test_plan_names_relink_keep_and_skip_rows(self):
+        code, out = self.dry_run()
+        self.assertEqual(code, 0, out)
+        items = {item["row_id"]: item for item in self.plan()["rows"]}
+        rows = self.rows
+        self.assertNotIn(rows["G"], items, "a PromptManager run is not foreign")
+        actions = {key: items[rows[key]]["action"] for key in rows if key != "G"}
+        self.assertEqual(
+            actions,
+            {
+                "A": "relink",
+                "B": "relink",
+                "C": "keep",
+                "D": "skip",
+                "E": "skip",
+                "F": "relink",
+                "F_ok": "keep",
+                "H": "relink",
+            },
+        )
+        a = items[rows["A"]]
+        self.assertEqual(a["current_prompt_id"], self.orphan)
+        self.assertEqual(a["current_text_sha256"], sha("an old guess"))
+        self.assertEqual(a["target_text"], TEXTS[1])
+        self.assertEqual(a["target_hash"], generate_prompt_hash(TEXTS[1]))
+        self.assertEqual(a["target_prompt_id"], "new")
+        self.assertEqual(a["file"], self.files["A"])
+        self.assertEqual(items[rows["B"]]["target_prompt_id"], self.p2)
+        self.assertEqual(items[rows["H"]]["target_prompt_id"], "new")
+        self.assertEqual(items[rows["D"]]["reason"], "file gone")
+        self.assertEqual(
+            items[rows["E"]]["reason"], "not decided by the recorded saver"
+        )
+        self.assertIn("relink: 4", out)
+        self.assertIn("keep: 2", out)
+        self.assertIn("skip: 2", out)
+
+    def test_orphan_audit_lists_the_prompts_losing_every_link(self):
+        self.dry_run()
+        plan = self.plan()
+        orphans = {o["id"]: o for o in plan["orphans"]}
+        self.assertEqual(set(orphans), {self.orphan, self.rated})
+        self.assertEqual(orphans[self.orphan]["blockers"], [])
+        self.assertTrue(orphans[self.orphan]["safe"])
+        self.assertEqual(orphans[self.rated]["blockers"], ["usage or rating"])
+        self.assertFalse(orphans[self.rated]["safe"])
+        self.assertFalse(plan["delete_orphans"], "off by default")
+        self.dry_run("--delete-orphans")
+        self.assertTrue(self.plan()["delete_orphans"])
+
+    def test_dry_run_writes_nothing_and_writes_the_markdown_list(self):
+        before = file_sha(self.db_path)
+        listing = sorted(os.listdir(self.tmp))
+        code, out = self.dry_run()
+        self.assertEqual(code, 0)
+        self.assertEqual(file_sha(self.db_path), before)
+        self.assertIn("Dry run: nothing was changed.", out)
+        # A read-only reader of a WAL library may leave its empty side files
+        sidecars = {"prompts.db-shm", "prompts.db-wal"}
+        self.assertEqual(
+            sorted(set(os.listdir(self.tmp)) - sidecars),
+            sorted(set(listing + ["plan.json", "plan.list.md"]) - sidecars),
+        )
+        wal = Path(self.tmp, "prompts.db-wal")
+        self.assertEqual(wal.stat().st_size if wal.exists() else 0, 0)
+        markdown = Path(self.tmp, "plan.list.md").read_text(encoding="utf-8")
+        self.assertIn(f"| {self.rows['A']} |", markdown)
+        long_path = os.path.join("dataset", NAME.format(1))
+        self.assertIn(f"| {long_path} |", markdown, "file paths are never cut")
+        self.assertIn(generate_prompt_hash(TEXTS[1])[:12], markdown)
+        self.assertIn("new", markdown)
+        self.assertIn("would lose every link", markdown)
+
+
+class DryRunReviewTests(RepairFixture):
+    def test_plan_records_its_library_and_fingerprint(self):
+        self.dry_run()
+        plan = self.plan()
+        self.assertEqual(
+            plan["source_db"], os.path.normcase(os.path.realpath(self.db_path))
+        )
+        self.assertEqual(plan["max_row_id"], max(self.links()))
+
+    def test_a_markdown_plan_path_is_refused(self):
+        out = io.StringIO()
+        with (
+            self.assertRaises(SystemExit),
+            contextlib.redirect_stderr(io.StringIO()),
+        ):
+            tool.main(
+                [
+                    self.db_path,
+                    "--output-dir",
+                    self.output,
+                    "--plan",
+                    os.path.join(self.tmp, "plan.md"),
+                ],
+                out=out,
+            )
+        self.assertFalse(os.path.exists(os.path.join(self.tmp, "plan.md")))
+
+    def test_keep_compares_the_linked_text_not_the_stored_hash(self):
+        self.query_write("UPDATE prompts SET hash='stale' WHERE id=?", (self.p0,))
+        self.dry_run()
+        items = {item["row_id"]: item for item in self.plan()["rows"]}
+        self.assertEqual(items[self.rows["C"]]["action"], "keep")
+
+    def test_delete_orphans_is_rehearsed_with_the_deletes(self):
+        code, out = self.dry_run("--delete-orphans")
+        self.assertEqual(code, 0, out)
+        simulated = self.plan()["simulated"]
+        self.assertEqual(simulated["result"]["orphans_deleted"], 1)
+        self.assertEqual(
+            simulated["after"]["prompts"], simulated["before"]["prompts"] + 2 - 1
+        )
+
+    def test_a_file_on_another_drive_is_listed_by_its_full_path(self):
+        self.dry_run()
+        plan = self.plan()
+        with unittest.mock.patch.object(
+            tool.os.path, "relpath", side_effect=ValueError("other drive")
+        ):
+            listing = tool.markdown(plan, {})
+        self.assertIn(self.files["A"], listing)
+
+
+class ApplyTests(RepairFixture):
+    def setUp(self):
+        super().setUp()
+        code, out = self.dry_run()
+        self.assertEqual(code, 0, out)
+
+    def test_apply_relinks_creates_targets_and_drops_pair_links(self):
+        code, out = self.apply()
+        self.assertEqual(code, 0, out)
+        links = self.links()
+        new = dict(self.query("SELECT hash, id FROM prompts"))
+        target_a = new[generate_prompt_hash(TEXTS[1])]
+        target_h = new[generate_prompt_hash(TEXTS[3])]
+        self.assertEqual(links[self.rows["A"]], target_a)
+        self.assertEqual(links[self.rows["B"]], self.p2)
+        self.assertEqual(links[self.rows["H"]], target_h)
+        self.assertEqual(links[self.rows["C"]], self.p0, "keep")
+        self.assertEqual(links[self.rows["E"]], self.p0, "skip")
+        self.assertEqual(links[self.rows["G"]], self.p0, "not foreign")
+        self.assertEqual(
+            self.query(
+                "SELECT COUNT(*) FROM image_prompt_pairs WHERE image_id=?",
+                (self.rows["B"],),
+            ),
+            [(0,)],
+        )
+        category, notes = self.query(
+            "SELECT category, notes FROM prompts WHERE id=?", (target_a,)
+        )[0]
+        self.assertEqual(category, "scanned")
+        self.assertEqual(
+            notes, f"Auto-scanned from {os.path.basename(self.files['A'])}"
+        )
+        tags = self.query(
+            "SELECT t.name FROM prompt_tags pt JOIN tags t ON t.id=pt.tag_id "
+            "WHERE pt.prompt_id=?",
+            (target_a,),
+        )
+        self.assertEqual(tags, [("auto-scanned",)])
+        self.assertIn("integrity_check", out)
+        self.assertIn("foreign_key_check", out)
+        self.assertRegex(out, r"integrity_check\s+ok\s+ok")
+        # Orphans are kept unless the plan chose to delete them
+        self.assertTrue(self.query("SELECT 1 FROM prompts WHERE id=?", (self.orphan,)))
+
+    def test_unique_case_deletes_the_wrong_row(self):
+        code, out = self.apply()
+        self.assertEqual(code, 0, out)
+        links = self.links()
+        self.assertNotIn(self.rows["F"], links)
+        self.assertEqual(links[self.rows["F_ok"]], self.p2)
+        rows = self.query(
+            "SELECT prompt_id FROM generated_images WHERE image_path=?",
+            (self.files["F"],),
+        )
+        self.assertEqual(rows, [(self.p2,)])
+
+    def test_backup_is_taken_before_writing(self):
+        before = self.links()
+        code, out = self.apply()
+        self.assertEqual(code, 0, out)
+        names = os.listdir(self.backups)
+        self.assertEqual(len(names), 1)
+        self.assertRegex(names[0], r"^prompts\.db\.pre-247-repair_\d{8}_\d{6}$")
+        backup = os.path.join(self.backups, names[0])
+        conn = sqlite3.connect(backup)
+        try:
+            saved = dict(conn.execute("SELECT id, prompt_id FROM generated_images"))
+        finally:
+            conn.close()
+        self.assertEqual(saved, before)
+        self.assertIn(f"Backup: {backup}", out)
+
+    def test_backup_beside_the_database_by_default(self):
+        out = io.StringIO()
+        code = tool.main(
+            [
+                self.db_path,
+                "--apply",
+                self.plan_path,
+                "--comfyui-dir",
+                os.path.join(self.tmp, "comfy"),
+            ],
+            out=out,
+            probe=not_running,
+            handles=no_handles,
+        )
+        self.assertEqual(code, 0, out.getvalue())
+        self.assertTrue(
+            any(
+                n.startswith("prompts.db.pre-247-repair_") for n in os.listdir(self.tmp)
+            )
+        )
+
+    def test_refuses_while_comfyui_runs(self):
+        before = file_sha(self.db_path)
+        code, out = self.apply(probe=lambda _d, _p: ["process 42 (main.py)"])
+        self.assertEqual(code, 2)
+        self.assertIn("REFUSED", out)
+        self.assertIn("process 42", out)
+        self.assertEqual(os.listdir(self.backups), [], "no backup either")
+        self.assertEqual(file_sha(self.db_path), before)
+
+    def test_rehearsal_skips_the_running_check_on_a_copy_only(self):
+        code, out = self.apply("--rehearsal", probe=lambda _d, _p: ["port 8188"])
+        self.assertEqual(code, 0, out)
+        self.assertIn("rehearsal", out)
+        # The live library of that ComfyUI folder: --rehearsal is refused
+        live_dir = os.path.join(self.tmp, "comfy", "user", "__promptmanager")
+        os.makedirs(live_dir)
+        live = os.path.join(live_dir, "prompts.db")
+        shutil.copy(self.db_path, live)
+        out = io.StringIO()
+        code = tool.main(
+            [
+                live,
+                "--apply",
+                self.plan_path,
+                "--comfyui-dir",
+                os.path.join(self.tmp, "comfy"),
+                "--rehearsal",
+            ],
+            out=out,
+            probe=lambda _d, _p: [],
+            handles=no_handles,
+        )
+        self.assertEqual(code, 2)
+        self.assertIn("--rehearsal", out.getvalue())
+
+    def test_drift_refuses_and_changes_nothing(self):
+        cases = {
+            "row moved": lambda: self.query_write(
+                "UPDATE generated_images SET prompt_id=? WHERE id=?",
+                (self.p2, self.rows["H"]),
+            ),
+            "text changed": lambda: self.query_write(
+                "UPDATE prompts SET text='an older guess' WHERE id=?", (self.orphan,)
+            ),
+            "file changed": lambda: png(self.files["B"], shapes.branch_bench()),
+            "target stored since": lambda: self.query_write(
+                "INSERT INTO prompts(text, hash) VALUES(?,?)",
+                (TEXTS[1], generate_prompt_hash(TEXTS[1])),
+            ),
+        }
+        pristine = os.path.join(self.tmp, "pristine.db")
+        shutil.copy(self.db_path, pristine)
+        image = Path(self.files["B"]).read_bytes()
+        for name, drift in cases.items():
+            with self.subTest(name):
+                shutil.copy(pristine, self.db_path)
+                Path(self.files["B"]).write_bytes(image)
+                shutil.rmtree(self.backups)
+                os.makedirs(self.backups)
+                drift()
+                before = dump(self.db_path)
+                code, out = self.apply()
+                self.assertEqual(code, 2, out)
+                self.assertIn("REFUSED (rolled back, nothing changed)", out)
+                self.assertEqual(dump(self.db_path), before)
+
+    def test_late_failure_rolls_back_every_write(self):
+        plan = self.plan()
+        plan["delete_orphans"] = True
+        self.write_plan(plan)
+        # The safe orphan gained a user tag after the dry run
+        self.query_write("INSERT INTO tags(name) VALUES('mine')", ())
+        self.query_write(
+            "INSERT INTO prompt_tags(prompt_id, tag_id) "
+            "SELECT ?, id FROM tags WHERE name='mine'",
+            (self.orphan,),
+        )
+        before = self.links()
+        prompts = self.query("SELECT COUNT(*) FROM prompts")
+        code, out = self.apply()
+        self.assertEqual(code, 2, out)
+        self.assertIn(f"prompt {self.orphan}", out)
+        self.assertIn("user tags", out)
+        self.assertEqual(self.links(), before, "the relinks were rolled back")
+        self.assertEqual(self.query("SELECT COUNT(*) FROM prompts"), prompts)
+
+    def test_orphans_are_deleted_only_when_chosen_and_unreferenced(self):
+        plan = self.plan()
+        plan["delete_orphans"] = True
+        self.write_plan(plan)
+        code, out = self.apply()
+        self.assertEqual(code, 0, out)
+        self.assertFalse(self.query("SELECT 1 FROM prompts WHERE id=?", (self.orphan,)))
+        self.assertTrue(
+            self.query("SELECT 1 FROM prompts WHERE id=?", (self.rated,)),
+            "a prompt with a rating is never deleted",
+        )
+        self.assertIn("orphans_deleted", out)
+
+    def test_an_applied_plan_refuses_a_second_time(self):
+        self.assertEqual(self.apply()[0], 0)
+        before = dump(self.db_path)
+        code, out = self.apply()
+        self.assertEqual(code, 2)
+        self.assertIn("REFUSED", out)
+        self.assertEqual(dump(self.db_path), before)
+
+
+class ApplySafetyTests(RepairFixture):
+    def setUp(self):
+        super().setUp()
+        code, out = self.dry_run()
+        self.assertEqual(code, 0, out)
+
+    def assert_unchanged(self, before, code, out):
+        self.assertEqual(code, 2, out)
+        self.assertEqual(dump(self.db_path), before)
+
+    def test_refuses_when_another_process_has_the_library_open(self):
+        before = dump(self.db_path)
+        held = lambda _db: ["process 7 has prompts.db-wal open"]  # noqa: E731
+        for args in ((), ("--rehearsal",)):
+            with self.subTest(args):
+                code, out = self.apply(*args, handles=held)
+                self.assert_unchanged(before, code, out)
+                self.assertIn("process 7", out)
+        self.assertEqual(os.listdir(self.backups), [])
+
+    def test_the_running_check_runs_again_just_before_writing(self):
+        calls = []
+
+        def late(_d, _p):
+            calls.append(1)
+            return [] if len(calls) == 1 else ["something listens on port 8188"]
+
+        before = dump(self.db_path)
+        code, out = self.apply(probe=late)
+        self.assert_unchanged(before, code, out)
+        self.assertEqual(len(calls), 2)
+        self.assertIn("rolled back", out)
+        self.assertIn("Backup:", out)
+
+    def test_without_proc_only_an_explicit_stop_lets_it_through(self):
+        blind = lambda _db: [tool.NO_PROC]  # noqa: E731
+        before = dump(self.db_path)
+        code, out = self.apply(handles=blind)
+        self.assert_unchanged(before, code, out)
+        self.assertIn("--i-stopped-comfyui", out)
+        code, out = self.apply("--i-stopped-comfyui", handles=blind)
+        self.assertEqual(code, 0, out)
+
+    def test_a_folder_without_main_py_is_refused(self):
+        os.remove(os.path.join(self.comfy, "main.py"))
+        before = dump(self.db_path)
+        code, out = self.apply()
+        self.assert_unchanged(before, code, out)
+        self.assertIn("main.py", out)
+
+    def test_a_plan_for_another_library_is_refused(self):
+        other = os.path.join(self.tmp, "other.db")
+        shutil.copy(self.db_path, other)
+        out = io.StringIO()
+        code = tool.main(
+            [other, "--apply", self.plan_path, "--comfyui-dir", self.comfy],
+            out=out,
+            probe=not_running,
+            handles=no_handles,
+        )
+        self.assertEqual(code, 2)
+        self.assertIn("made for", out.getvalue())
+
+    def test_opening_the_library_must_not_change_its_schema(self):
+        self.query_write("DROP INDEX idx_prompt_pairs_positive_only", ())
+        PromptModel.reset_schema_cache()
+        code, out = self.apply()
+        self.assertEqual(code, 2, out)
+        self.assertIn("schema", out)
+        self.assertIn("live code version", out)
+
+    def test_an_existing_backup_file_is_refused(self):
+        stamp = "20261028_000000"
+        Path(self.backups, f"prompts.db.pre-247-repair_{stamp}").write_bytes(b"x")
+        before = dump(self.db_path)
+        with unittest.mock.patch.object(tool, "_stamp", return_value=stamp):
+            code, out = self.apply()
+        self.assert_unchanged(before, code, out)
+        self.assertIn("already exists", out)
+
+    def test_a_backup_that_does_not_match_is_refused(self):
+        real = tool.backup_database
+
+        def short(source, target):
+            real(source, target)
+            conn = sqlite3.connect(target)
+            try:
+                with conn:
+                    conn.execute("DELETE FROM image_prompt_pairs")
+            finally:
+                conn.close()
+
+        before = dump(self.db_path)
+        with unittest.mock.patch.object(tool, "backup_database", short):
+            code, out = self.apply()
+        self.assert_unchanged(before, code, out)
+        self.assertIn("backup", out)
+
+    def test_a_changed_target_text_is_refused(self):
+        plan = self.plan()
+        for item in plan["rows"]:
+            if item["row_id"] == self.rows["A"]:
+                item["target_text"] = "something else"
+        self.write_plan(plan)
+        before = dump(self.db_path)
+        code, out = self.apply()
+        self.assert_unchanged(before, code, out)
+        self.assertIn("target_text", out)
+
+    def test_an_orphan_gone_or_changed_at_apply_is_refused(self):
+        plan = self.plan()
+        plan["delete_orphans"] = True
+        plan["orphans"][0]["text_sha256"] = "0" * 64
+        self.write_plan(plan)
+        before = dump(self.db_path)
+        code, out = self.apply()
+        self.assert_unchanged(before, code, out)
+        self.assertIn("gone or text changed", out)
+
+    def test_any_error_rolls_back_and_names_the_backup(self):
+        before = dump(self.db_path)
+        with unittest.mock.patch.object(
+            tool, "_relink", side_effect=RuntimeError("disk on fire")
+        ):
+            code, out = self.apply()
+        self.assert_unchanged(before, code, out)
+        self.assertIn("rolled back", out)
+        self.assertIn("disk on fire", out)
+        self.assertIn(self.backups, out)
+
+    def test_a_failed_integrity_check_after_writing_exits_1(self):
+        real = tool.counts
+        seen = []
+
+        def counts(path, readonly=True):
+            found = real(path, readonly)
+            if path == self.db_path:
+                seen.append(path)
+                if len(seen) > 1:
+                    return {**found, "integrity_check": "corrupt"}
+            return found
+
+        with unittest.mock.patch.object(tool, "counts", counts):
+            code, out = self.apply()
+        self.assertEqual(code, 1, out)
+        self.assertIn("restore", out)
+
+
+class BlockerTests(RepairFixture):
+    """Each reference that keeps a prompt from being deleted."""
+
+    def test_every_blocker(self):
+        pid = self.orphan
+        cases = {
+            "linked images": None,
+            "saved pairs": "INSERT INTO prompt_pairs(positive_prompt_id) VALUES(:id)",
+            "recorded runs": "UPDATE execution_run_contexts "
+            "SET positive_prompt_id=:id",
+            "library state": "INSERT INTO prompt_library_state(prompt_id) "
+            "VALUES(:id)",
+            "user tags": "INSERT INTO prompt_tags(prompt_id, tag_id) "
+            "SELECT :id, id FROM tags WHERE name='mine'",
+            "usage or rating": "UPDATE prompts SET run_count=3 WHERE id=:id",
+            "category": "UPDATE prompts SET category='portraits' WHERE id=:id",
+            "notes": "UPDATE prompts SET notes='mine' WHERE id=:id",
+        }
+        conn = sqlite3.connect(self.db_path)
+        try:
+            conn.execute("INSERT INTO tags(name) VALUES('mine')")
+            conn.commit()
+            for name, sql in cases.items():
+                with self.subTest(name):
+                    conn.execute("SAVEPOINT blocker")
+                    if sql is None:
+                        unused = {pid}
+                    else:
+                        conn.execute(
+                            "DELETE FROM generated_images WHERE prompt_id=?", (pid,)
+                        )
+                        conn.execute(sql, {"id": pid})
+                        unused = {pid}
+                    found = tool.blockers(conn, pid, unused)
+                    self.assertTrue(any(item.startswith(name) for item in found), found)
+                    conn.execute("ROLLBACK TO blocker")
+                    conn.execute("RELEASE blocker")
+            conn.execute("DELETE FROM generated_images WHERE prompt_id=?", (pid,))
+            self.assertEqual(tool.blockers(conn, pid, {pid}), [])
+            self.assertEqual(
+                tool.blockers(conn, pid, set()), ["not unused by the app's rule"]
+            )
+            conn.rollback()
+        finally:
+            conn.close()
+
+
+def _off_text_node(text, save=False):
+    return {
+        "class_type": "PromptManagerText",
+        "inputs": {"text": text, "save_to_library": save},
+    }
+
+
+def _encode_from(source):
+    return {
+        "class_type": "CLIPTextEncode",
+        "inputs": {"text": source, "clip": ["1", 1]},
+    }
+
+
+def off_behind_pass_through(branch=2):
+    """The dataset graph with ``branch``'s text coming from a switched-off
+    PromptManagerText through a pass-through the #90 walk does not follow."""
+    graph = shapes.shared_prefix()
+    graph["50"] = _off_text_node(TEXTS[branch - 1])
+    graph["51"] = {"class_type": "MyNode", "inputs": {"value": ["50", 0]}}
+    graph[str(branch * 10)] = _encode_from(["51", 0])
+    return graph
+
+
+def off_target(branch=2, save=False):
+    """``branch``'s text straight from a PromptManagerText whose switch is
+    ``save`` (False, or a link to a node the graph cannot read)."""
+    graph = shapes.shared_prefix()
+    graph["50"] = _off_text_node(TEXTS[branch - 1], save)
+    graph[str(branch * 10)] = _encode_from(["50", 0])
+    if isinstance(save, list):
+        graph["52"] = {"class_type": "SomeBoolNode", "inputs": {}}
+    return graph
+
+
+def off_negative_only():
+    """Every branch's negative from a switched-off PromptManagerText."""
+    graph = shapes.shared_prefix()
+    graph["50"] = _off_text_node(shapes.NEGATIVE)
+    graph["3"] = _encode_from(["50", 0])
+    return graph
+
+
+class SaveToLibraryTests(RecordedRuns, unittest.TestCase):
+    """#90: the repair never relinks a file to a switched-off node's text.
+
+    The plan decides each file through ``decide_file_prompt``; every answer
+    but a positive from the recorded saver or the inspector is a skip, so a
+    ``not_saved`` or ``not_saved_wired`` file is never written. Each file is
+    reported by a run without contexts (as a cached or unrecorded run is),
+    so only the ladder can stop it. Saver "23" is branch 2.
+    """
+
+    def setUp(self):
+        self.tmp = tempfile.mkdtemp()
+        self.addCleanup(shutil.rmtree, self.tmp, True)
+        self.addCleanup(assert_closed, self, self.tmp)
+        self.addCleanup(PromptModel.reset_schema_cache)
+        self.output = os.path.join(self.tmp, "output")
+        os.makedirs(os.path.join(self.output, "dataset"))
+        self.db_path = os.path.join(self.tmp, "prompts.db")
+
+    def plan_for(self, graph):
+        path = os.path.join(self.output, "dataset", NAME.format(1))
+        png(path, graph)
+        db = PromptDatabase(self.db_path)
+        try:
+            wrong = db.save_prompt(
+                text=TEXTS[0], prompt_hash=generate_prompt_hash(TEXTS[0])
+            )
+            db.link_image_to_prompt(wrong, path, metadata={})
+            self.record_run(
+                db, self.output, [("dataset", NAME.format(1))], [], nodes=["23"]
+            )
+            plan = tool.build_plan(db, self.output, self.db_path)
+        finally:
+            db.close_all()
+        (item,) = plan["rows"]
+        return item
+
+    def test_an_off_node_behind_a_pass_through_is_skipped(self):
+        item = self.plan_for(off_behind_pass_through())
+        self.assertEqual(item["action"], "skip")
+        self.assertEqual(item["decision"], "not_saved/none")
+        self.assertNotIn("target_text", item)
+
+    def test_an_off_node_in_the_recorded_run_target_is_skipped(self):
+        cases = {
+            "off": (False, "not_saved/none"),
+            "wired": (["52", 0], "not_saved_wired/none"),
+        }
+        for name, (save, decision) in cases.items():
+            with self.subTest(name):
+                if os.path.exists(self.db_path):
+                    os.remove(self.db_path)
+                item = self.plan_for(off_target(save=save))
+                self.assertEqual(item["action"], "skip")
+                self.assertEqual(item["decision"], decision)
+
+    def test_an_off_node_on_another_branch_still_skips_the_file(self):
+        """The ladder judges the file's whole graph: any switched-off text
+        upstream of a positive makes the file not saved."""
+        item = self.plan_for(off_target(branch=3))
+        self.assertEqual(item["action"], "skip")
+        self.assertEqual(item["decision"], "not_saved/none")
+
+    def test_an_off_node_wired_only_to_the_negative_still_repairs(self):
+        item = self.plan_for(off_negative_only())
+        self.assertEqual(item["action"], "relink")
+        self.assertEqual(item["source"], "recorded-run")
+        self.assertEqual(item["target_text"], TEXTS[1])
+
+    def test_a_switched_on_node_in_the_target_still_repairs(self):
+        """Regression guard: a switch that is on is no reason to skip."""
+        item = self.plan_for(off_target(save=True))
+        self.assertEqual(item["action"], "relink")
+        self.assertEqual(item["target_text"], TEXTS[1])
+
+
+class ActionTests(unittest.TestCase):
+    """Only the recorded saver relinks; a certain step naming the linked
+    prompt keeps it; the legacy floor and the rest are skipped."""
+
+    def test_action_by_decision(self):
+        linked = generate_prompt_hash("linked")
+        cases = [
+            (Decision("positive", "other", source="recorded-run"), "relink"),
+            (Decision("positive", "linked", source="recorded-run"), "keep"),
+            (Decision("positive", "linked", source="inspector"), "keep"),
+            (Decision("positive", "other", source="inspector"), "skip"),
+            (Decision("positive", "linked", source="legacy-parser"), "skip"),
+            (Decision("pair", "linked", source="resolver"), "skip"),
+            (Decision("never"), "skip"),
+            (Decision("not_saved"), "skip"),
+            (Decision("not_saved_wired"), "skip"),
+            (Decision("positive", "other", source="opted-out-negative"), "skip"),
+        ]
+        for decision, action in cases:
+            with self.subTest(decision.source, text=decision.text):
+                self.assertEqual(tool.action_of(decision, linked), action)
+
+
+class CommandLineTests(unittest.TestCase):
+    def test_running_probe_reports_a_listening_port(self):
+        import socket
+
+        with socket.socket() as server:
+            server.bind(("127.0.0.1", 0))
+            server.listen(1)
+            port = server.getsockname()[1]
+            found = tool.comfyui_running("/nonexistent/comfy", port)
+        self.assertTrue(any(str(port) in item for item in found))
+
+    def test_running_probe_is_quiet_when_nothing_runs(self):
+        import socket
+
+        with socket.socket() as probe:
+            probe.bind(("127.0.0.1", 0))
+            port = probe.getsockname()[1]
+        self.assertEqual(tool.comfyui_running("/nonexistent/comfy", port), [])
+
+    def fake_proc(self, tmp):
+        proc = Path(tmp, "proc")
+        db = Path(tmp, "prompts.db")
+        db.write_bytes(b"")
+        Path(tmp, "prompts.db-wal").write_bytes(b"")
+        comfy = Path(tmp, "comfy")
+        comfy.mkdir()
+        holder = proc / "125" / "fd"
+        holder.mkdir(parents=True)
+        os.symlink(Path(tmp, "prompts.db-wal"), holder / "3")
+        (proc / "125" / "cmdline").write_bytes(b"python\0main.py\0")
+        os.symlink(comfy, proc / "125" / "cwd")
+        (proc / "123").mkdir()  # unreadable: no fd, cmdline or cwd
+        (proc / "self").mkdir()  # not a pid
+        return proc, db, comfy
+
+    @unittest.skipIf(os.name == "nt", "symlinks need privileges on Windows")
+    def test_proc_scan_finds_holders_and_skips_unreadable_entries(self):
+        with tempfile.TemporaryDirectory() as tmp:
+            proc, db, comfy = self.fake_proc(tmp)
+            found = tool.open_by_others(str(db), proc=str(proc), own_pid=1)
+            self.assertEqual(len(found), 1)
+            self.assertIn("125", found[0])
+            self.assertEqual(
+                tool.open_by_others(str(db), proc=str(proc), own_pid=125), []
+            )
+            running = tool.comfyui_running(str(comfy), 1, proc=str(proc))
+            self.assertTrue(any("125" in item for item in running), running)
+
+    def test_without_proc_the_scan_says_it_cannot_tell(self):
+        self.assertEqual(
+            tool.open_by_others("/x/prompts.db", proc="/nonexistent/proc"),
+            [tool.NO_PROC],
+        )
+
+    def test_missing_database_is_an_argument_error(self):
+        with self.assertRaises(SystemExit), contextlib.redirect_stderr(io.StringIO()):
+            tool.main(
+                ["/nonexistent/prompts.db", "--output-dir", "/tmp"], out=io.StringIO()
+            )
+
+    def test_apply_needs_the_comfyui_folder(self):
+        with tempfile.TemporaryDirectory() as tmp:
+            db = os.path.join(tmp, "prompts.db")
+            Path(db).write_bytes(b"")
+            with (
+                self.assertRaises(SystemExit),
+                contextlib.redirect_stderr(io.StringIO()),
+            ):
+                tool.main([db, "--apply", "plan.json"], out=io.StringIO())
+
+
+if __name__ == "__main__":
+    unittest.main()

tests/test_run_attribution.py

@@ -33,6 +33,7 @@ from utils.execution_outputs import (
     snapshot_output_roots,
 )
 from utils.hashing import generate_prompt_hash
+from py.config import GalleryConfig
 from utils.image_monitor import ImageGenerationHandler
 
 STYLE = "windows" if os.name == "nt" else "posix"
@@ -771,6 +772,487 @@ class TestMonitorHistoryRecheck(RecordedRuns, unittest.TestCase):
         run_attribution.notify_outputs_sealed("run", [])  # no dead callback raises
 
 
+def folder_is_case_sensitive(folder):
+    """True when ``folder``'s volume tells names apart by case: the existing
+    folder spelled with swapped case is not found (#476)."""
+    head, name = os.path.split(folder)
+    return not os.path.exists(os.path.join(head, name.swapcase()))
+
+
+class TestMonitorForeignRuns(RecordedRuns, unittest.TestCase):
+    """#247: outputs of workflows without a PromptManager node, in the monitor."""
+
+    def setUp(self):
+        history_tests.HistoryDatabaseTests.setUp(self)
+        self.db = history_tests.HistoryDatabaseTests.database(self)
+        self.output = str(Path(self.temp.name) / "output")
+        os.makedirs(os.path.join(self.output, "dataset"))
+        self.tracker = make_tracker(self.db)
+        self.handler = ImageGenerationHandler(self.db, self.tracker)
+        self.handler.processing_delay = 0
+        self.handler.settle_interval = 0.01
+        self.addCleanup(close_handler, self, self.handler)
+        self.metadata = {"prompt": shapes.shared_prefix()}
+        saved = GalleryConfig.SAVE_FOREIGN_RUNS
+        self.addCleanup(setattr, GalleryConfig, "SAVE_FOREIGN_RUNS", saved)
+        GalleryConfig.SAVE_FOREIGN_RUNS = True
+        patches = (
+            mock.patch(
+                "utils.image_monitor.host_output_directory", return_value=self.output
+            ),
+            mock.patch(
+                "database.run_attribution.host_output_directory",
+                return_value=self.output,
+            ),
+            mock.patch.object(
+                self.handler.metadata_extractor,
+                "extract_metadata",
+                side_effect=lambda path: self.metadata,
+            ),
+        )
+        for patch in patches:
+            patch.start()
+            self.addCleanup(patch.stop)
+
+    close_databases = history_tests.HistoryDatabaseTests.close_databases
+    wait_for = TestMonitorHistoryRecheck.wait_for
+
+    def image(self, name, subfolder="dataset"):
+        path = os.path.join(self.output, subfolder, name)
+        Path(path).write_bytes(b"png")
+        return path
+
+    def linked_text(self, path):
+        row = (
+            self.db.model.get_connection()
+            .execute(
+                "SELECT p.text FROM generated_images g "
+                "JOIN prompts p ON p.id = g.prompt_id WHERE g.image_path = ?",
+                (str(path),),
+            )
+            .fetchone()
+        )
+        return row[0] if row else None
+
+    def key(self, path):
+        return self.handler._awaiting_key(path)
+
+    def prompt_count(self):
+        connection = self.db.model.get_connection()
+        return connection.execute("SELECT COUNT(*) FROM prompts").fetchone()[0]
+
+    def settled(self):
+        return self.wait_for(lambda: self.handler.pending_count() == 0)
+
+    def test_each_dataset_file_links_to_its_branch_once_the_run_ends(self):
+        """CUJ 1: unlinked while the run is open, its own branch after the seal."""
+        paths = [self.image(f"IMG_GEN_0000{n}_.png") for n in (1, 2)]
+        for path in paths:
+            self.handler.process_new_image(path)
+        self.assertEqual([self.linked_text(path) for path in paths], [None, None])
+        self.assertEqual(
+            self.handler.awaiting_foreign(), [self.key(path) for path in paths]
+        )
+        self.assertEqual(self.handler.awaiting_history(), [])
+        self.record_run(
+            self.db,
+            self.output,
+            [("dataset", os.path.basename(path)) for path in paths],
+            [],
+            nodes=["13", "23"],
+        )
+        self.assertTrue(self.wait_for(lambda: all(map(self.linked_text, paths))))
+        self.assertTrue(self.settled())
+        self.assertEqual(
+            [self.linked_text(path) for path in paths], list(shapes.BRANCH_TEXTS[:2])
+        )
+        self.assertEqual(self.handler.awaiting_foreign(), [])
+        prompt = self.db.get_prompt_by_hash(
+            generate_prompt_hash(shapes.BRANCH_TEXTS[0])
+        )
+        self.assertEqual(prompt["category"], "scanned")
+        self.assertIn("Saved from a workflow run: IMG_GEN_00001_.png", prompt["notes"])
+
+    def test_a_run_sealed_before_the_file_settles_links_it_at_once(self):
+        path = self.image("IMG_GEN_00003_.png")
+        self.record_run(
+            self.db, self.output, [("dataset", "IMG_GEN_00003_.png")], [], nodes=["33"]
+        )
+        self.handler.process_new_image(path)
+        self.assertEqual(self.linked_text(path), shapes.BRANCH_TEXTS[2])
+        self.assertEqual(self.handler.awaiting_foreign(), [])
+
+    def test_a_name_two_runs_reported_stays_unlinked(self):
+        """CUJ 3: the file was overwritten; no run proves its saver."""
+        path = self.image("IMG_GEN_00004_.png")
+        reported = [("dataset", "IMG_GEN_00004_.png")]
+        self.record_run(self.db, self.output, reported, [], nodes=["13"])
+        self.record_run(self.db, self.output, reported, [], item=1, nodes=["23"])
+        self.handler.process_new_image(path)
+        self.assertIsNone(self.linked_text(path))
+        reference = {"folder_type": "output", "subfolder": "dataset"}
+        reference["filename"] = "IMG_GEN_00004_.png"
+        with self.assertLogs("prompt_manager.image_monitor", "INFO") as logs:
+            self.handler._recheck_foreign(path, reference)
+        self.assertIn(
+            "No certain prompt for IMG_GEN_00004_.png after its run ended; "
+            "left unlinked",
+            "\n".join(logs.output),
+        )
+        self.assertIsNone(self.linked_text(path))
+        self.assertEqual(self.prompt_count(), 0)
+
+    def test_the_setting_off_stores_nothing_at_settle_or_after_the_run(self):
+        """CUJ 4: read for every file, so it applies without a restart."""
+        GalleryConfig.SAVE_FOREIGN_RUNS = False
+        skipped = self.image("IMG_GEN_00005_.png")
+        with self.assertLogs("prompt_manager.image_monitor", "INFO") as logs:
+            self.handler.process_new_image(skipped)
+        self.assertIn(
+            "No PromptManager prompt in workflow, skipping image",
+            "\n".join(logs.output),
+        )
+        self.assertEqual(self.handler.awaiting_foreign(), [])
+        GalleryConfig.SAVE_FOREIGN_RUNS = True
+        waiting = self.image("IMG_GEN_00006_.png")
+        self.handler.process_new_image(waiting)
+        self.assertEqual(self.handler.awaiting_foreign(), [self.key(waiting)])
+        GalleryConfig.SAVE_FOREIGN_RUNS = False
+        self.record_run(
+            self.db, self.output, [("dataset", "IMG_GEN_00006_.png")], [], nodes=["13"]
+        )
+        self.assertTrue(self.wait_for(lambda: not self.handler.awaiting_foreign()))
+        self.assertTrue(self.settled())
+        self.assertEqual(
+            [self.linked_text(p) for p in (skipped, waiting)], [None, None]
+        )
+        self.assertEqual(self.prompt_count(), 0)
+
+    def test_a_plain_foreign_workflow_links_at_settle_and_counts_no_run(self):
+        """CUJ 2: the default profile decides; the tracker queue is untouched."""
+        text = "a lighthouse on a cliff at dusk"
+        self.metadata = {"prompt": sampled_graph(text, "blurry")}
+        queued = self.db.save_prompt(
+            text="queued item", prompt_hash=generate_prompt_hash("queued item")
+        )
+        self.tracker.set_current_prompt("queued item", {"prompt_id": queued})
+        first, second = self.image("plain_00001_.png"), self.image("plain_00002_.png")
+        self.handler.process_new_image(first)
+        self.handler.process_new_image(second)
+        self.assertEqual(
+            [self.linked_text(first), self.linked_text(second)], [text, text]
+        )
+        prompt = self.db.get_prompt_by_hash(generate_prompt_hash(text))
+        self.assertEqual(prompt["run_count"] or 0, 0)
+        self.assertIsNone(prompt.get("last_used_at"))
+        self.assertEqual(self.tracker.pop_next_prompt()["id"], queued)
+        self.assertEqual(self.handler.awaiting_foreign(), [])
+
+    def test_the_sealed_reference_spells_the_file_as_history_recorded_it(self):
+        """Windows: the watcher's spelling differs in case from the run's report.
+
+        On a case-sensitive folder the two spellings are two files and the
+        folded key joins them; on a case-insensitive one (Windows, default
+        macOS) they already name one file, so only history's spelling and
+        the watcher's differ (#476)."""
+        seen = os.path.join(self.output, "dataset", "img_gen_00007_.png")
+        Path(seen).write_bytes(b"png")
+        recorded = os.path.join(self.output, "Dataset", "IMG_GEN_00007_.png")
+        self.assertNotEqual(os.path.basename(recorded), os.path.basename(seen))
+        if folder_is_case_sensitive(os.path.join(self.output, "dataset")):
+            Path(self.output, "Dataset").mkdir()
+            self.image("IMG_GEN_00007_.png", subfolder="Dataset")
+        else:
+            self.assertTrue(os.path.exists(recorded))
+        folded = staticmethod(lambda path: os.path.realpath(path).lower())
+        with mock.patch.object(ImageGenerationHandler, "_awaiting_key", folded):
+            self.handler.process_new_image(seen)
+            self.assertIsNone(self.linked_text(seen))
+            self.record_run(
+                self.db,
+                self.output,
+                [("Dataset", os.path.basename(recorded))],
+                [],
+                nodes=["43"],
+            )
+            self.assertTrue(self.wait_for(lambda: self.linked_text(seen)))
+            self.assertTrue(self.settled())
+        self.assertEqual(self.linked_text(seen), shapes.BRANCH_TEXTS[3])
+
+    def test_a_full_foreign_set_never_evicts_a_waiting_keyless_file(self):
+        from utils import image_monitor
+
+        keyless = self.image("bare_00001_.png", subfolder="")
+        with mock.patch.object(
+            self.handler.metadata_extractor, "extract_metadata", return_value=None
+        ):
+            self.handler.process_new_image(keyless)
+        paths = [self.image(f"IMG_GEN_1000{n}_.png") for n in range(3)]
+        with (
+            mock.patch.object(image_monitor, "MAX_AWAITING_HISTORY", 2),
+            self.assertLogs("prompt_manager.image_monitor", "INFO") as logs,
+        ):
+            for path in paths:
+                self.handler.process_new_image(path)
+            self.assertEqual(
+                self.handler.awaiting_foreign(), [self.key(p) for p in paths[1:]]
+            )
+        self.assertEqual(self.handler.awaiting_history(), [self.key(keyless)])
+        self.assertIn(
+            "Too many files wait for their run; IMG_GEN_10000_.png left unlinked",
+            "\n".join(logs.output),
+        )
+
+    def test_an_expired_foreign_entry_is_logged_and_left_unlinked(self):
+        from utils import image_monitor
+
+        path = self.image("IMG_GEN_00008_.png")
+        self.handler.process_new_image(path)
+        later = time.monotonic() + image_monitor.AWAIT_HISTORY_SECONDS + 1
+        with (
+            mock.patch.object(image_monitor.time, "monotonic", return_value=later),
+            self.assertLogs("prompt_manager.image_monitor", "INFO") as logs,
+        ):
+            self.assertEqual(self.handler.awaiting_foreign(), [])
+        self.assertIn(
+            "No recorded run for IMG_GEN_00008_.png; left unlinked",
+            "\n".join(logs.output),
+        )
+        self.assertIsNone(self.linked_text(path))
+
+    def test_a_seal_while_the_pending_set_is_full_is_rechecked_later(self):
+        from utils import image_monitor
+
+        path = self.image("IMG_GEN_00009_.png")
+        self.handler.process_new_image(path)
+        blocker = self.image("blocker.png", subfolder="")
+        release = threading.Event()
+        original = self.handler.process_new_image
+
+        def slow(image_path, prompt_snapshot=None):
+            if image_path == blocker:
+                release.wait(5)
+                return None
+            return original(image_path, prompt_snapshot=prompt_snapshot)
+
+        self.handler.process_new_image = slow
+        with mock.patch.object(image_monitor, "MAX_PENDING_IMAGES", 1):
+            self.assertTrue(self.handler.schedule(blocker))
+            self.assertTrue(self.wait_for(lambda: self.handler.pending_count() == 1))
+            self.record_run(
+                self.db,
+                self.output,
+                [("dataset", "IMG_GEN_00009_.png")],
+                [],
+                nodes=["13"],
+            )
+            self.assertEqual(list(self.handler._foreign_due), [path])
+            release.set()
+            self.assertTrue(self.wait_for(lambda: self.linked_text(path)))
+        self.assertTrue(self.settled())
+        self.assertEqual(self.linked_text(path), shapes.BRANCH_TEXTS[0])
+
+    def test_shutdown_forgets_waiting_and_due_foreign_files(self):
+        path = self.image("IMG_GEN_00010_.png")
+        self.handler.process_new_image(path)
+        self.handler._foreign_due[path] = {"filename": "IMG_GEN_00010_.png"}
+        self.handler.shutdown()
+        self.assertEqual(self.handler.awaiting_foreign(), [])
+        self.assertEqual(list(self.handler._foreign_due), [])
+
+    def link_count(self, path):
+        connection = self.db.model.get_connection()
+        return connection.execute(
+            "SELECT COUNT(*) FROM generated_images WHERE image_path = ?", (str(path),)
+        ).fetchone()[0]
+
+    def test_a_file_linked_to_another_prompt_never_gets_a_second_link(self):
+        """One prompt per file: the pair path and the positive path both refuse."""
+        other = self.db.save_prompt(
+            "an earlier prompt", generate_prompt_hash("an earlier prompt")
+        )
+        plain = self.image("plain_00004_.png")
+        self.db.link_image_to_prompt(other, plain)
+        self.metadata = {"prompt": sampled_graph("a quiet harbor", "blurry")}
+        with self.assertLogs("prompt_manager.image_monitor", "INFO") as logs:
+            self.handler.process_new_image(plain)
+        self.assertIn("Foreign run: conflict", "\n".join(logs.output))
+        dataset = self.image("IMG_GEN_00011_.png")
+        self.db.link_image_to_prompt(other, dataset)
+        self.metadata = {"prompt": shapes.shared_prefix()}
+        self.record_run(
+            self.db, self.output, [("dataset", "IMG_GEN_00011_.png")], [], nodes=["13"]
+        )
+        self.handler.process_new_image(dataset)
+        for path in (plain, dataset):
+            self.assertEqual(self.linked_text(path), "an earlier prompt")
+            self.assertEqual(self.link_count(path), 1)
+        for text in ("a quiet harbor", shapes.BRANCH_TEXTS[0]):
+            self.assertIsNone(self.db.get_prompt_by_hash(generate_prompt_hash(text)))
+
+    def test_the_file_size_is_read_only_when_the_ladder_needs_it(self):
+        """PR 1's lazy size: a foreign file decided without a store reads none."""
+        # Foreign (a core encoder only) and no API graph: never, at step 3
+        encoder = {"type": "CLIPTextEncode", "inputs": [], "widgets_values": ["x"]}
+        self.metadata = {"text_encoder_nodes": [encoder]}
+        path = self.image("bare_00002_.png", subfolder="")
+        with mock.patch("utils.foreign_runs.os.path.getsize") as size:
+            self.handler.process_new_image(path)
+        size.assert_not_called()
+
+    def untraceable(self):
+        """Unresolved and only the legacy parser names a text (an unknown node)."""
+        graph = sampled_graph("unused", "blurry")
+        graph["1"]["inputs"]["text"] = ["30", 0]
+        graph["30"] = {
+            "class_type": "VendorPromptMaker",
+            "inputs": {"prompt": "a misty valley at dawn"},
+        }
+        return {"prompt": graph}
+
+    def test_no_record_at_settle_keeps_the_file_waiting_for_the_legacy_floor(self):
+        """Review: a lookup that found no record is not the run's answer yet."""
+        self.metadata = self.untraceable()
+        path = self.image("vendor_00001_.png")
+        self.handler.process_new_image(path)
+        self.assertIsNone(self.linked_text(path))
+        self.assertEqual(self.handler.awaiting_foreign(), [self.key(path)])
+        self.record_run(self.db, self.output, [("dataset", "vendor_00001_.png")], [])
+        self.assertTrue(self.wait_for(lambda: self.linked_text(path)))
+        self.assertTrue(self.settled())
+        self.assertEqual(self.linked_text(path), "a misty valley at dawn")
+
+    def test_a_record_found_at_settle_without_a_certain_prompt_ends_the_wait(self):
+        """Review: sealed before settle, still ambiguous: no seal is coming."""
+        path = self.image("IMG_GEN_00012_.png")
+        # One run reported it from two savers: a record, but no saver
+        reported = [("dataset", "IMG_GEN_00012_.png")] * 2
+        self.record_run(self.db, self.output, reported, [], nodes=["13", "23"])
+        with self.assertLogs("prompt_manager.image_monitor", "INFO") as logs:
+            self.handler.process_new_image(path)
+        self.assertIn(
+            "No certain prompt for IMG_GEN_00012_.png; left unlinked",
+            "\n".join(logs.output),
+        )
+        self.assertEqual(self.handler.awaiting_foreign(), [])
+        self.assertIsNone(self.linked_text(path))
+
+    def test_monitoring_saved_off_stops_foreign_writes_at_settle_and_recheck(self):
+        """Review: the observer runs until a restart; nothing is written."""
+        enabled = GalleryConfig.MONITORING_ENABLED
+        self.addCleanup(setattr, GalleryConfig, "MONITORING_ENABLED", enabled)
+        GalleryConfig.MONITORING_ENABLED = False
+        self.metadata = {"prompt": sampled_graph("a quiet harbor", "blurry")}
+        plain = self.image("plain_00005_.png")
+        self.handler.process_new_image(plain)
+        self.assertIsNone(self.linked_text(plain))
+        GalleryConfig.MONITORING_ENABLED = True
+        self.metadata = {"prompt": shapes.shared_prefix()}
+        waiting = self.image("IMG_GEN_00013_.png")
+        self.handler.process_new_image(waiting)
+        GalleryConfig.MONITORING_ENABLED = False
+        reference = {"folder_type": "output", "subfolder": "dataset"}
+        reference["filename"] = "IMG_GEN_00013_.png"
+        self.record_run(
+            self.db, self.output, [("dataset", reference["filename"])], [], nodes=["13"]
+        )
+        self.assertTrue(self.wait_for(lambda: not self.handler.awaiting_foreign()))
+        self.assertTrue(self.settled())
+        self.handler._recheck_foreign(waiting, reference)
+        self.assertIsNone(self.linked_text(waiting))
+        self.assertEqual(self.prompt_count(), 0)
+
+    def test_the_setting_is_read_again_just_before_the_store(self):
+        self.metadata = {"prompt": sampled_graph("a quiet harbor", "blurry")}
+        path = self.image("plain_00006_.png")
+        real = __import__("utils.foreign_runs", fromlist=["x"]).decide_foreign
+
+        def decide_then_switch_off(*args, **kwargs):
+            decision = real(*args, **kwargs)
+            GalleryConfig.SAVE_FOREIGN_RUNS = False
+            return decision
+
+        with mock.patch(
+            "utils.image_monitor.decide_foreign", side_effect=decide_then_switch_off
+        ):
+            self.handler.process_new_image(path)
+        self.assertIsNone(self.linked_text(path))
+        self.assertEqual(self.prompt_count(), 0)
+
+    def test_a_seal_between_registering_and_the_settle_lookup_links_once(self):
+        """Regression guard (passes before the review fixes): one link only."""
+        path = self.image("IMG_GEN_00014_.png")
+        real = __import__("utils.foreign_runs", fromlist=["x"]).history_lookup
+
+        def sealed_first(db, image_path, reference=None):
+            lookup = real(db, image_path, reference)
+
+            def run_then_look():
+                if reference is None:  # the settle lookup: the run seals now
+                    self.record_run(
+                        self.db,
+                        self.output,
+                        [("dataset", "IMG_GEN_00014_.png")],
+                        [],
+                        nodes=["13"],
+                    )
+                return lookup()
+
+            return run_then_look
+
+        with mock.patch("utils.image_monitor.history_lookup", side_effect=sealed_first):
+            self.assertTrue(self.handler.schedule(path))
+            self.assertTrue(self.wait_for(lambda: self.linked_text(path)))
+            self.assertTrue(self.settled())
+        self.assertEqual(self.linked_text(path), shapes.BRANCH_TEXTS[0])
+        self.assertEqual(self.link_count(path), 1)
+        self.assertEqual(list(self.handler._foreign_due), [])
+
+    def test_a_drained_recheck_never_drops_a_newer_reference(self):
+        path = self.image("IMG_GEN_00015_.png")
+        older, newer = {"filename": "old"}, {"filename": "new"}
+        self.handler._foreign_due[path] = older
+        real = self.handler.schedule
+
+        def schedule(image_path, snapshot=None):
+            # The run is sealed again while the older one is scheduled
+            self.handler._foreign_due[image_path] = newer
+            return real(image_path, snapshot)
+
+        self.handler.schedule = schedule
+        with mock.patch.object(self.handler, "process_new_image"):
+            self.handler._drain_history_due()
+        self.assertIs(self.handler._foreign_due[path], newer)
+
+    def test_an_evicted_due_recheck_is_logged(self):
+        from utils import image_monitor
+
+        self.handler.schedule = lambda *args, **kwargs: False
+        with (
+            mock.patch.object(image_monitor, "MAX_AWAITING_HISTORY", 1),
+            self.assertLogs("prompt_manager.image_monitor", "INFO") as logs,
+        ):
+            for name in ("a_00001_.png", "b_00001_.png"):
+                self.handler._schedule_recheck(self.image(name), {"filename": name})
+        self.assertIn(
+            "Too many files wait for their run; a_00001_.png left unlinked",
+            "\n".join(logs.output),
+        )
+
+    def test_a_failed_store_is_logged_and_leaves_the_file_unlinked(self):
+        self.metadata = {"prompt": sampled_graph("a quiet harbor", "blurry")}
+        path = self.image("plain_00003_.png")
+        with (
+            mock.patch(
+                "utils.foreign_runs.ingest_resolved_pair",
+                side_effect=sqlite3.OperationalError("locked"),
+            ),
+            self.assertLogs("prompt_manager.image_monitor", "ERROR"),
+        ):
+            self.handler.process_new_image(path)
+        self.assertIsNone(self.linked_text(path))
+
+
 class TestBoundedLookup(RecordedRuns, unittest.TestCase):
     """M5: history lookups are batched, and the history schema is unchanged."""
 

tests/test_saved_views.py

@@ -42,7 +42,14 @@ IMAGES_STATE = {
     "favorites": False,
     "collection": "",
 }
-LIBRARY_STATE = {"q": "", "category": "", "sort": "rating", "tags": []}
+LIBRARY_STATE = {
+    "q": "",
+    "category": "",
+    "sort": "rating",
+    "tags": [],
+    "tag_mode": "untagged",
+    "folder": "2026/Aug",
+}
 IMAGES_DEFAULTS = {
     "q": "",
     "category": "",
@@ -54,7 +61,14 @@ IMAGES_DEFAULTS = {
     "favorites": False,
     "collection": "",
 }
-LIBRARY_DEFAULTS = {"q": "", "category": "", "sort": "newest", "tags": []}
+LIBRARY_DEFAULTS = {
+    "q": "",
+    "category": "",
+    "sort": "newest",
+    "tags": [],
+    "tag_mode": "all",
+    "folder": "",
+}
 # A row saved by an older build: no favorites key, a retired key, a sort
 # that is no longer offered and an over-long folder.
 OLD_IMAGES_STATE = json.dumps(
@@ -228,9 +242,17 @@ class TestSavedViewsApi(ImageAPITestCase):
         favorite = await self.create(state={"favorites": True})
         self.assertIs(favorite["state"]["favorites"], True)
         for sort in ("newest", "oldest", "name", "rating"):
-            await self.create(page="library", state={"sort": sort, "tags": ["a b"]})
+            await self.create(
+                page="library",
+                state={"sort": sort, "tags": ["a b"], "tag_mode": "all", "folder": ""},
+            )
+        for mode in ("untagged", "any"):
+            view = await self.create(
+                page="library", state={"tags": [], "tag_mode": mode}
+            )
+            self.assertEqual(view["state"]["tag_mode"], mode)
         _, listed = await self.send("GET")
-        self.assertEqual(len(listed["views"]), 20)
+        self.assertEqual(len(listed["views"]), 22)
 
     async def test_write_requires_the_collection_key(self):
         # #331 D140: a view saved on a collection page keeps the collection id.
@@ -675,6 +697,32 @@ class TestSavedViewsSchema(unittest.TestCase):
                 )
         self.assertEqual(normalize_state("history", {}), {})
 
+    def test_library_tag_mode_and_folder_are_kept_with_defaults_for_old_rows(self):
+        # A row from before #444 has neither key.
+        old = {"q": "", "category": "", "sort": "rating", "tags": ["a"]}
+        self.assertEqual(normalize_state("library", old), {**LIBRARY_DEFAULTS, **old})
+        self.assertEqual(
+            normalize_state("library", {"tag_mode": "or", "folder": 3}),
+            LIBRARY_DEFAULTS,
+        )
+        for mode in ("all", "any", "untagged"):
+            self.assertEqual(
+                normalize_state("library", {"tag_mode": mode})["tag_mode"], mode
+            )
+        with self.assertRaises(SavedViewInvalid):
+            encode_state("library", {**LIBRARY_STATE, "tag_mode": "or"})
+        with self.assertRaises(SavedViewInvalid):
+            encode_state("library", {**LIBRARY_STATE, "folder": "x" * 1025})
+
+    def test_library_untagged_is_never_paired_with_tags(self):
+        # Never written by this build: the tags win, as the listing refuses both.
+        self.assertEqual(
+            normalize_state("library", {"tags": ["a"], "tag_mode": "untagged"}),
+            {**LIBRARY_DEFAULTS, "tags": ["a"]},
+        )
+        with self.assertRaises(SavedViewInvalid):
+            encode_state("library", {**LIBRARY_STATE, "tags": ["a"]})
+
     def test_validation_errors_are_saved_view_invalid(self):
         for call in (
             lambda: validate_name(""),

tests/test_settings_sources.py

@@ -323,6 +323,43 @@ class TestMonitoringSetting(SourcesTestCase):
         self.assertIs(GalleryConfig.MONITORING_ENABLED, False)
 
 
+class TestSaveForeignRunsSetting(SourcesTestCase):
+    """#247 D-Q4: "Save prompts from other workflows", applied live."""
+
+    async def asyncSetUp(self):
+        await super().asyncSetUp()
+        saved = GalleryConfig.SAVE_FOREIGN_RUNS
+        self.addCleanup(setattr, GalleryConfig, "SAVE_FOREIGN_RUNS", saved)
+        GalleryConfig.SAVE_FOREIGN_RUNS = True
+
+    async def test_read_and_write_published_live_without_a_restart(self):
+        self.assertIs((await self._get())["settings"]["save_foreign_runs"], True)
+        self._install_monitor(FakeMonitor([self.output_dir]))
+        status, data = await self._post(
+            "/prompt_manager/settings", {"save_foreign_runs": False}
+        )
+        self.assertEqual(status, 200, data)
+        self.assertTrue(data["persisted"])
+        self.assertEqual(data["restart_required_fields"], [])
+        saved = self._read_config()
+        self.assertIs(saved["gallery"]["monitoring"]["save_foreign_runs"], False)
+        self.assertIs(GalleryConfig.SAVE_FOREIGN_RUNS, False)
+        self.assertIs((await self._get())["settings"]["save_foreign_runs"], False)
+
+    async def test_only_true_or_false_is_accepted(self):
+        self._write_config({"custom": {"keep": True}})
+        before = self._config_bytes()
+        for value in ("false", 0, None, []):
+            with self.subTest(value=value):
+                status, data = await self._post(
+                    "/prompt_manager/settings", {"save_foreign_runs": value}
+                )
+                self.assertEqual(status, 400, data)
+                self.assertIn("save_foreign_runs must be true or false", data["error"])
+                self.assertEqual(self._config_bytes(), before)
+                self.assertIs(GalleryConfig.SAVE_FOREIGN_RUNS, True)
+
+
 class TestSourceChangeHold(SourcesTestCase):
     async def test_root_save_refused_while_scan_job_runs(self):
         GalleryConfig.MONITORING_DIRECTORIES = [self._root(self.a)]

tests/test_vocabulary.py

@@ -364,6 +364,162 @@ class TestVocabularyReads(unittest.TestCase):
             self.assertTrue(conn.in_transaction)
 
 
+class _ReadFixture(unittest.TestCase):
+    """TestVocabularyReads' database and helpers, without re-running its tests."""
+
+    setUp = TestVocabularyReads.setUp
+    tearDown = TestVocabularyReads.tearDown
+    save = TestVocabularyReads.save
+    read = TestVocabularyReads.read
+    add_unused = TestVocabularyReads.add_unused
+    all_tags = TestVocabularyReads.all_tags
+
+
+class TestVocabularyTagSearchAndSort(_ReadFixture):
+    """#444: tag search, tag sort and the untagged prompts count."""
+
+    def names(self, **options):
+        rows, after = [], None
+        while True:
+            page = self.read("list_vocabulary_tags", limit=2, after=after, **options)
+            rows.extend(row["name"] for row in page["tags"])
+            if not page["has_more"]:
+                return rows
+            after = page["next_key"]
+
+    def seed_counts(self):
+        self.save("One", tags=["beta", "Alpha"])
+        self.save("Two", tags=["beta", "gamma"])
+        self.save("Three", tags=["beta"])
+        self.add_unused("delta", "100%_off", "100x-off")
+
+    def test_search_is_a_case_insensitive_literal_substring(self):
+        self.seed_counts()
+        self.assertEqual(self.names(search="ALP"), ["Alpha"])
+        self.assertEqual(self.names(search="%"), ["100%_off"])
+        self.assertEqual(self.names(search="_"), ["100%_off"])
+        self.assertEqual(
+            self.names(search="  a  "), ["beta", "Alpha", "gamma", "delta"]
+        )
+        page = self.read("list_vocabulary_tags", search="a")
+        self.assertEqual(page["total"], 4)
+        self.assertEqual(self.read("list_vocabulary_tags", search="")["total"], 6)
+
+    def test_each_sort_orders_rows_and_continues_without_gaps(self):
+        self.seed_counts()
+        self.assertEqual(
+            self.names(), ["beta", "Alpha", "gamma", "delta", "100%_off", "100x-off"]
+        )
+        self.assertEqual(self.names(sort="created"), self.names())
+        self.assertEqual(
+            self.names(sort="name"),
+            ["100%_off", "100x-off", "Alpha", "beta", "delta", "gamma"],
+        )
+        self.assertEqual(
+            self.names(sort="name-desc"),
+            ["gamma", "delta", "beta", "Alpha", "100x-off", "100%_off"],
+        )
+        # Ties keep creation order: Alpha and gamma (1), then the unused tags (0).
+        self.assertEqual(
+            self.names(sort="prompts"),
+            ["beta", "Alpha", "gamma", "delta", "100%_off", "100x-off"],
+        )
+        self.assertEqual(
+            self.names(sort="prompts-asc"),
+            ["delta", "100%_off", "100x-off", "Alpha", "gamma", "beta"],
+        )
+        self.assertEqual(
+            self.names(sort="name", search="a"), ["Alpha", "beta", "delta", "gamma"]
+        )
+
+    def test_count_sort_uses_the_row_counts(self):
+        self.save("Parent", tags=["direct"])
+        self.db.save_prompt_pair("Owner", "Noise", tags=["pair"])
+        self.db.save_prompt_pair("Owner two", "Noise two", tags=["pair"])
+        archived = self.db.save_prompt_pair("Gone", "Gone noise", tags=["archived"])
+        self.db.archive_prompt_pair(archived)
+        page = self.read("list_vocabulary_tags", sort="prompts")
+        self.assertEqual(
+            [(row["name"], row["prompt_count"]) for row in page["tags"]],
+            [("pair", 2), ("direct", 1), ("archived", 0)],
+        )
+
+    def test_invalid_search_sort_and_keys_are_refused(self):
+        self.seed_counts()
+        for options in (
+            {"sort": "count"},
+            {"sort": None},
+            {"search": "x" * 201},
+            {"search": 3},
+            {"sort": "name", "after": 3},
+            {"sort": "name", "after": ["a"]},
+            {"sort": "prompts", "after": ["a", 1]},
+            {"sort": "created", "after": ["a", 1]},
+        ):
+            with self.assertRaises(ValueError, msg=options):
+                self.read("list_vocabulary_tags", **options)
+
+    def test_untagged_count_is_every_parent_no_tag_browse_returns(self):
+        self.save("Plain")
+        self.save("Tagged", tags=["t"])
+        self.db.save_prompt_pair("Pair tagged", "Noise", tags=["p"])
+        self.db.save_prompt_pair("Pair plain", "Noise plain")
+        archived = self.db.save_prompt_pair("Archived tag", "Old", tags=["old"])
+        self.db.archive_prompt_pair(archived)
+        self.add_unused("unused")
+        page = self.read("list_vocabulary_tags", search="zzz")
+        untagged = self.db.list_library(tag_mode="untagged")
+        self.assertEqual(
+            sorted(item["text"] for item in untagged["items"]),
+            ["Archived tag", "Pair plain", "Plain"],
+        )
+        self.assertEqual(page["untagged_count"], untagged["total"])
+        self.assertEqual(page["untagged_count"], 3)
+        tagged = set()
+        for row in self.all_tags():
+            tagged |= {
+                item["id"] for item in self.db.list_library(tags=[row["name"]])["items"]
+            }
+        everyone = {item["id"] for item in self.db.list_library()["items"]}
+        self.assertEqual({item["id"] for item in untagged["items"]}, everyone - tagged)
+
+
+class TestLibraryTagMode(_ReadFixture):
+    def test_untagged_mode_combines_with_text_and_category(self):
+        self.save("Red plain", category="Colors")
+        self.save("Blue plain")
+        self.save("Red tagged", category="Colors", tags=["red"])
+        self.assertEqual(
+            [
+                i["text"]
+                for i in self.db.list_library(text="red", tag_mode="untagged")["items"]
+            ],
+            ["Red plain"],
+        )
+        self.assertEqual(
+            [
+                i["text"]
+                for i in self.db.list_library(category="Colors", tag_mode="untagged")[
+                    "items"
+                ]
+            ],
+            ["Red plain"],
+        )
+        self.assertEqual(
+            self.db.list_library(tag_mode="all")["total"],
+            self.db.list_library()["total"],
+        )
+
+    def test_untagged_mode_refuses_tags_and_unknown_modes(self):
+        for options in (
+            {"tag_mode": "untagged", "tags": ["x"]},
+            {"tag_mode": "none"},
+            {"tag_mode": None},
+        ):
+            with self.assertRaises(ValueError, msg=options):
+                self.db.list_library(**options)
+
+
 _TAG_TABLES = ('"tags"', '"prompt_tags"', '"prompt_pair_tags"', '"sqlite_sequence"')
 
 

tests/vendor/ci-workflow.test.mjs

@@ -212,11 +212,15 @@ test('every browser job sets PM_TEST_WORKERS and runs in the pinned image', () =
     assert.match(mergeJob(), PINNED_IMAGE, 'the merge sees the same workspace paths as the shards');
 });
 
-test('page workflows run each browser in two shards', () => {
+// #473: two shards split by test count put the slow concept specs in shard 1, which reached the 18-minute step limit.
+test('page workflows run each browser in three shards', () => {
     const workflows = allJobs.get('browser-workflows');
     assert.deepEqual(matrixList(workflows, 'browser'), ['firefox', 'webkit']);
-    assert.deepEqual(matrixList(workflows, 'shard'), ['1', '2']);
-    assert.match(workflows, /test:workflows -- --project \$\{\{ matrix\.browser \}\} --shard \$\{\{ matrix\.shard \}\}\/2/);
+    assert.deepEqual(matrixList(workflows, 'shard'), ['1', '2', '3']);
+    assert.match(workflows, /test:workflows -- --project \$\{\{ matrix\.browser \}\} --shard \$\{\{ matrix\.shard \}\}\/3/);
+    assert.match(workflows, /^ {4}name: page workflows \(\$\{\{ matrix\.browser \}\}, shard \$\{\{ matrix\.shard \}\}\/3\)$/m);
+    assert.match(workflows, /name: Page workflows in \$\{\{ matrix\.browser \}\}, shard \$\{\{ matrix\.shard \}\}\/3$/m);
+    assert.doesNotMatch(workflows, /\/2\b/, 'no step still names two shards');
     assert.match(workflows, /^ {6}fail-fast: false$/m);
     assert.match(workflows, /name: browser-traces-\$\{\{ matrix\.browser \}\}-shard-\$\{\{ matrix\.shard \}\}/);
 });

tests/vendor/comfyui-config.test.mjs

@@ -36,14 +36,14 @@ test('execution script discovers only the new matrix and preserves all 78 origin
     // + the fresh-host library location case, + #342 image attribution, + #343 linked text,
     // + #332 image category routes on a generated output (image-attribution.spec.mjs, 2 projects),
     // + #347 cached runtime-linked runs: five node kinds and eight one-renderer cases (execution-text-link.spec.mjs, 2 projects),
-    // + #90 Save to library off (five node types)
-    assert.match(executions, /Total: 128 tests in 5 files/);
+    // + #90 Save to library off (five node types), + #247 foreign workflows (image-attribution.spec.mjs, 2 tests x 2 projects)
+    assert.match(executions, /Total: 132 tests in 5 files/);
     for (const project of ['classic-start', 'nodes2-start']) {
         const cases = executions.split('\n').filter(line => line.includes(`[${project}]`));
         assert.equal(cases.filter(line => line.includes('execution-history.spec.mjs')).length, 15);
         assert.equal(cases.filter(line => line.includes('execution-outputs.spec.mjs')).length, 8);
         assert.equal(cases.filter(line => line.includes('library-location.spec.mjs')).length, 1);
-        assert.equal(cases.filter(line => line.includes('image-attribution.spec.mjs')).length, 3);
+        assert.equal(cases.filter(line => line.includes('image-attribution.spec.mjs')).length, 5);
         assert.equal(cases.filter(line => line.includes('execution-text-link.spec.mjs')).length, 37);
     }
     assert.doesNotMatch(executions, /prompt-manager.spec/);

tools/comfyui_stopped.py

@@ -0,0 +1,81 @@
+"""Is ComfyUI stopped? The checks a library repair runs before writing (#247).
+
+``comfyui_running`` looks for ``main.py`` processes of one ComfyUI folder
+and a listener on its port. ``open_by_others`` lists the processes that hold
+the library file, its ``-wal`` or its ``-shm`` open. Both read ``/proc``
+(Linux); without it ``open_by_others`` answers ``[NO_PROC]``: it cannot
+tell, and the caller refuses unless the user says ComfyUI is stopped.
+Entries that cannot be read (another user's process, one that just exited)
+are skipped.
+"""
+
+import os
+import socket
+
+NO_PROC = (
+    "cannot list the processes that hold the library open (no /proc): stop "
+    "ComfyUI and pass --i-stopped-comfyui"
+)
+
+
+def _same(path):
+    return os.path.normcase(os.path.realpath(path))
+
+
+def _pids(proc, own_pid=None):
+    try:
+        names = os.listdir(proc)
+    except OSError:
+        return []
+    return [name for name in names if name.isdigit() and name != str(own_pid)]
+
+
+def comfyui_running(comfyui_dir, port, proc="/proc"):
+    """``main.py`` processes running from ``comfyui_dir`` and a listener on
+    ``port`` (127.0.0.1), each as one line; empty when none is found."""
+    found = []
+    folder = _same(comfyui_dir)
+    for pid in _pids(proc):
+        try:
+            with open(os.path.join(proc, pid, "cmdline"), "rb") as stream:
+                parts = stream.read().split(b"\0")
+            cwd = _same(os.readlink(os.path.join(proc, pid, "cwd")))
+        except OSError:
+            continue
+        if any(part.endswith(b"main.py") for part in parts) and (
+            cwd == folder or any(folder.encode() in part for part in parts)
+        ):
+            found.append(f"process {pid} (main.py in {cwd})")
+    with socket.socket() as probe:
+        probe.settimeout(1)
+        if probe.connect_ex(("127.0.0.1", port)) == 0:
+            found.append(f"something listens on port {port}")
+    return found
+
+
+def open_by_others(database, proc="/proc", own_pid=None):
+    """Processes other than ``own_pid`` (default: this one) holding
+    ``database``, its ``-wal`` or ``-shm`` open; ``[NO_PROC]`` without /proc."""
+    if not os.path.isdir(proc):
+        return [NO_PROC]
+    own_pid = os.getpid() if own_pid is None else own_pid
+    targets = {_same(database + suffix): suffix for suffix in ("", "-wal", "-shm")}
+    found = []
+    for pid in _pids(proc, own_pid):
+        fd_dir = os.path.join(proc, pid, "fd")
+        try:
+            fds = os.listdir(fd_dir)
+        except OSError:
+            continue
+        held = set()
+        for fd in fds:
+            try:
+                target = os.readlink(os.path.join(fd_dir, fd))
+            except OSError:
+                continue
+            target = target.replace(" (deleted)", "")
+            if os.path.isabs(target) and _same(target) in targets:
+                held.add(os.path.basename(database) + targets[_same(target)])
+        if held:
+            found.append(f"process {pid} has {', '.join(sorted(held))} open")
+    return found

tools/repair_foreign_links.py

@@ -0,0 +1,776 @@
+"""Repair the links the old scan gave foreign dataset files (#247), on review.
+
+Before #247 the output scan linked each file of a foreign run (a workflow
+with no PromptManager node) through the legacy parser, so every file of a
+shared-prefix dataset run got the first branch's text. Recorded history
+knows which save node wrote each file; the shared ladder
+(``utils.file_prompt.decide_file_prompt``, scan policy) uses it
+(``recorded-run``) to name the right branch. This tool judges every
+``generated_images`` row of a file whose recording runs all have no
+PromptManager context:
+
+``relink``  the ladder names a positive by the recorded saver, and its text
+            is not the linked prompt's
+``keep``    the recorded saver or the inspector names the linked prompt
+``skip``    anything else (file gone, no single recorded saver, ambiguous,
+            the legacy floor or another step decided)
+
+Dry run (the default) judges a copy of the database (SQLite backup from a
+read-only connection) and writes a JSON plan and a Markdown list beside it
+(``<plan stem>.list.md``). It also rehearses the apply on a second copy, to
+list the prompts that would lose every link with their remaining
+references. Deleting those is OFF unless the dry run is given
+``--delete-orphans``; the choice is recorded in the plan and rehearsed on a
+third copy. The library is never written.
+
+``--apply <plan.json>`` (ComfyUI stopped), only on the library the plan was
+made from (its real path, recorded in the plan):
+
+1. refuses unless ``--comfyui-dir`` holds ``main.py``; refuses while a
+   ``main.py`` runs from it, anything listens on ``--port``, or any other
+   process holds the library, its ``-wal`` or ``-shm`` open
+   (``tools/comfyui_stopped.py``; without /proc only with
+   ``--i-stopped-comfyui``). ``--rehearsal`` skips the process and port
+   check, never the open-file check, and never for that folder's library;
+2. saves ``<db>.pre-247-repair_<stamp>`` (or into ``--backup-dir``), never
+   over an existing file, and refuses unless the backup passes
+   ``integrity_check`` with the library's counts. Opening the library must
+   not change its schema (a newer checkout's migration): run the tool from a
+   checkout at the live code version. The running check runs once more just
+   before writing;
+3. in ONE ``BEGIN IMMEDIATE`` transaction: re-judges every planned row and
+   refuses on any drift (row moved, prompt text changed, file changed, target
+   stored since), creates the target prompts that do not exist (category
+   ``scanned``, tag ``auto-scanned``, note ``Auto-scanned from <first file>``),
+   points each row at its target and drops its pair links (they named the old
+   positive). A file that already has a row for the target keeps that row and
+   loses the wrong one (UNIQUE(prompt_id, file_path)). Orphans are deleted only
+   when the plan chose it and each is still unreferenced.
+
+Any failed check rolls everything back and exits 2. Before/after table
+counts, ``integrity_check`` and ``foreign_key_check`` are printed.
+
+    python3 tools/repair_foreign_links.py <db> --output-dir <ComfyUI output> \\
+        --plan repair_plan.json
+    python3 tools/repair_foreign_links.py <db> --apply repair_plan.json \\
+        --comfyui-dir <ComfyUI folder>
+
+Run from a checkout at the live code version (after the merge: main) with
+its venv.
+"""
+
+import argparse
+import datetime
+import hashlib
+import json
+import logging
+import os
+import sqlite3
+import sys
+import tempfile
+from pathlib import Path
+from types import SimpleNamespace
+
+_ROOT = Path(__file__).resolve().parents[1]
+if str(_ROOT) not in sys.path:
+    sys.path.insert(0, str(_ROOT))
+
+from database.maintenance import ORPHANED_PROMPTS_SQL, _delete_ids  # noqa: E402
+from database.operations import PromptDatabase  # noqa: E402
+from database.pair_ingestion import (  # noqa: E402
+    SCAN_MARKS,
+    file_link_rows,
+    mark_created,
+)
+from database.run_attribution import recorded_output  # noqa: E402
+from py.api import PromptManagerAPI  # noqa: E402
+from tools.comfyui_stopped import (  # noqa: E402
+    NO_PROC,
+    comfyui_running,
+    open_by_others,
+)
+from tools.review_image_links import backup_database  # noqa: E402
+from utils.execution_outputs import snapshot_output_roots  # noqa: E402
+from utils.file_prompt import POSITIVE, Policy, decide_file_prompt  # noqa: E402
+from utils.hashing import generate_prompt_hash  # noqa: E402
+
+PLAN_VERSION = 1
+COMFY_PORT = 8188
+LIVE_LIBRARY = ("user", "__promptmanager", "prompts.db")
+EXCERPT = 80
+COUNTED = (
+    "prompts",
+    "generated_images",
+    "prompt_tags",
+    "image_prompt_pairs",
+    "workflow_documents",
+    "prompt_pairs",
+    "prompt_library_state",
+    "execution_run_contexts",
+    "tags",
+)
+# The ladder steps whose answer is certain enough to keep a link.
+KEEP_SOURCES = ("recorded-run", "inspector")
+# What must be unchanged since the dry run for a row to be written.
+DRIFT_KEYS = (
+    "action",
+    "file",
+    "current_prompt_id",
+    "current_text_sha256",
+    "file_size",
+    "file_mtime_ns",
+    "target_text",
+    "target_hash",
+    "target_prompt_id",
+)
+# Locations reported only by runs without any PromptManager context.
+_FOREIGN_LOCATIONS = (
+    "SELECT o.subfolder, o.filename FROM execution_run_outputs o "
+    "WHERE o.folder_type = 'output' AND o.root_key = ? AND o.path_style = ? "
+    "GROUP BY o.subfolder, o.filename "
+    "HAVING SUM(EXISTS (SELECT 1 FROM execution_run_contexts c "
+    "WHERE c.run_id = o.run_id)) = 0 "
+    "ORDER BY o.subfolder, o.filename"
+)
+
+
+class Refused(Exception):
+    """A check failed: nothing is written."""
+
+
+def sha256(text):
+    return hashlib.sha256(text.encode("utf-8")).hexdigest()
+
+
+def metadata_reader():
+    """The scan's metadata extraction (PNG text chunks, ffprobe for videos)."""
+    api = SimpleNamespace(logger=logging.getLogger("prompt_manager.repair_247"))
+    return lambda path: PromptManagerAPI._extract_comfyui_metadata(api, path) or {}
+
+
+def foreign_rows(db, output_dir):
+    """The link rows of every file only foreign runs reported, by id."""
+    style = "windows" if os.name == "nt" else "posix"
+    root_key = snapshot_output_roots(style, output_dir, output_dir).output_key
+    conn = db.model.get_connection()
+    rows = {}
+    for subfolder, filename in conn.execute(_FOREIGN_LOCATIONS, (root_key, style)):
+        path = os.path.join(output_dir, subfolder, filename)
+        rows.update((row["id"], row) for row in file_link_rows(conn, path))
+    return [rows[key] for key in sorted(rows)]
+
+
+class Judge:
+    """One file's decision per path, for all of its rows."""
+
+    def __init__(self, db, output_dir, read=None):
+        self.db = db
+        self.output_dir = output_dir
+        self.read = read or metadata_reader()
+        self.decisions = {}
+
+    def decide(self, path, size):
+        if path not in self.decisions:
+            self.decisions[path] = decide_file_prompt(
+                self.read(path),
+                file_name=os.path.basename(path),
+                file_size=size,
+                in_output=True,
+                lookup=lambda: recorded_output(self.db, path, self.output_dir),
+                policy=Policy.SCAN,
+            )
+        return self.decisions[path]
+
+    def row(self, row):
+        """The plan entry of one ``generated_images`` row."""
+        conn = self.db.model.get_connection()
+        path = row["image_path"]
+        (text,) = conn.execute(
+            "SELECT text FROM prompts WHERE id = ?", (row["prompt_id"],)
+        ).fetchone()
+        prompt_hash = generate_prompt_hash(text)
+        item = {
+            "row_id": row["id"],
+            "file": path,
+            "current_prompt_id": row["prompt_id"],
+            "current_text_sha256": sha256(text),
+        }
+        try:
+            stat = os.stat(path)
+        except OSError:
+            return {**item, "action": "skip", "reason": "file gone"}
+        item.update(file_size=stat.st_size, file_mtime_ns=stat.st_mtime_ns)
+        decision = self.decide(path, stat.st_size)
+        action = action_of(decision, prompt_hash)
+        if action == "skip":
+            return {
+                **item,
+                "action": "skip",
+                "reason": "not decided by the recorded saver",
+                "decision": f"{decision.kind}/{decision.source}",
+            }
+        target_hash = generate_prompt_hash(decision.text)
+        stored = conn.execute(
+            "SELECT id FROM prompts WHERE hash = ?", (target_hash,)
+        ).fetchone()
+        item.update(
+            target_text=decision.text,
+            target_hash=target_hash,
+            target_prompt_id=stored[0] if stored else "new",
+            action=action,
+            source=decision.source,
+        )
+        return item
+
+
+def action_of(decision, linked_hash):
+    """``relink``, ``keep`` or ``skip`` for a row linked to ``linked_hash``.
+
+    Only the recorded saver (``recorded-run``) names a new target. A certain
+    step (the recorded saver or the inspector) naming the linked prompt keeps
+    it; the legacy floor and every other answer are skipped.
+    """
+    if decision.kind != POSITIVE or decision.source not in KEEP_SOURCES:
+        return "skip"
+    if generate_prompt_hash(decision.text) == linked_hash:
+        return "keep"
+    return "relink" if decision.source == "recorded-run" else "skip"
+
+
+def library_identity(path):
+    """How a plan names its library: the case-normalized real path."""
+    return os.path.normcase(os.path.realpath(path))
+
+
+def build_plan(db, output_dir, source_db, read=None):
+    """The plan for ``db`` (a copy of ``source_db``)."""
+    judge = Judge(db, output_dir, read)
+    conn = db.model.get_connection()
+    (max_row_id,) = conn.execute("SELECT MAX(id) FROM generated_images").fetchone()
+    return {
+        "version": PLAN_VERSION,
+        "created": datetime.datetime.now().isoformat(timespec="seconds"),
+        "source_db": library_identity(source_db),
+        "max_row_id": max_row_id,
+        "output_dir": output_dir,
+        "rows": [judge.row(row) for row in foreign_rows(db, output_dir)],
+        "orphans": [],
+        "delete_orphans": False,
+    }
+
+
+def blockers(conn, prompt_id, unused):
+    """Why ``prompt_id`` must not be deleted (empty: nothing references it)."""
+    found = []
+    values = {"id": prompt_id, "tag": SCAN_MARKS.tags[0]}
+    checks = {
+        "linked images": "SELECT COUNT(*) FROM generated_images WHERE prompt_id=:id",
+        "saved pairs": "SELECT COUNT(*) FROM prompt_pairs "
+        "WHERE positive_prompt_id=:id OR negative_prompt_id=:id",
+        "recorded runs": "SELECT COUNT(*) FROM execution_run_contexts "
+        "WHERE positive_prompt_id=:id OR negative_prompt_id=:id",
+        "library state": "SELECT COUNT(*) FROM prompt_library_state "
+        "WHERE prompt_id=:id",
+        "user tags": "SELECT COUNT(*) FROM prompt_tags pt JOIN tags t "
+        "ON t.id=pt.tag_id WHERE pt.prompt_id=:id AND t.name != :tag",
+    }
+    for name, sql in checks.items():
+        count = conn.execute(sql, values).fetchone()[0]
+        if count:
+            found.append(f"{name}: {count}")
+    run_count, last_used, rating, category, notes = conn.execute(
+        "SELECT run_count, last_used_at, rating, category, notes FROM prompts "
+        "WHERE id=?",
+        (prompt_id,),
+    ).fetchone()
+    if run_count or last_used or rating:
+        found.append("usage or rating")
+    if category not in (None, SCAN_MARKS.category):
+        found.append(f"category {category}")
+    if notes and not notes.startswith("Auto-scanned from "):
+        found.append("notes")
+    if prompt_id not in unused:
+        found.append("not unused by the app's rule")
+    return found
+
+
+def _rejudge(conn, judge, plan):
+    for item in plan["rows"]:
+        if item["action"] == "skip":
+            continue
+        row = conn.execute(
+            "SELECT * FROM generated_images WHERE id = ?", (item["row_id"],)
+        ).fetchone()
+        if row is None:
+            raise Refused(f"row {item['row_id']}: gone")
+        if generate_prompt_hash(item.get("target_text") or "") != item.get(
+            "target_hash"
+        ):
+            raise Refused(f"row {item['row_id']}: target_text does not match its hash")
+        fresh = judge.row(row)
+        for key in DRIFT_KEYS:
+            if fresh.get(key) != item.get(key):
+                raise Refused(
+                    f"row {item['row_id']}: {key} changed "
+                    f"({item.get(key)!r} -> {fresh.get(key)!r})"
+                )
+
+
+def _targets(conn, db, relinks):
+    """``{target hash: prompt id}``, creating each "new" target once."""
+    targets = {}
+    for item in relinks:
+        digest = item["target_hash"]
+        if digest in targets:
+            continue
+        if item["target_prompt_id"] != "new":
+            targets[digest] = item["target_prompt_id"]
+            continue
+        prompt_id, created = db._pair_component_outcome(conn, item["target_text"])
+        if not created:
+            raise Refused(f"row {item['row_id']}: target prompt already stored")
+        mark_created(conn, db, prompt_id, SCAN_MARKS, os.path.basename(item["file"]))
+        targets[digest] = prompt_id
+    return targets
+
+
+def _relink(conn, item, target):
+    """Point the row at ``target``; True when the wrong row was deleted instead."""
+    row_id = item["row_id"]
+    correct = [
+        row
+        for row in file_link_rows(conn, item["file"])
+        if row["prompt_id"] == target and row["id"] != row_id
+    ]
+    if correct:  # UNIQUE(prompt_id, file_path): the correct row stays alone
+        if _delete_ids(conn, "generated_images", [row_id]) != 1:
+            raise Refused(f"row {row_id}: delete touched no row")
+        return True
+    # A pair link records the old positive: it goes with the old link
+    conn.execute("DELETE FROM image_prompt_pairs WHERE image_id = ?", (row_id,))
+    cursor = conn.execute(
+        "UPDATE generated_images SET prompt_id = ? WHERE id = ? AND prompt_id = ?",
+        (target, row_id, item["current_prompt_id"]),
+    )
+    if cursor.rowcount != 1:
+        raise Refused(f"row {row_id}: relink touched {cursor.rowcount} rows")
+    return False
+
+
+def _audit(conn, relinks):
+    """The prompts the relinks leave without any link, with their blockers."""
+    sources = {}
+    for item in relinks:
+        sources.setdefault(item["current_prompt_id"], []).append(item)
+    unused = {row[0] for row in conn.execute(ORPHANED_PROMPTS_SQL)}
+    audit = []
+    for prompt_id in sorted(sources):
+        if conn.execute(
+            "SELECT 1 FROM generated_images WHERE prompt_id = ?", (prompt_id,)
+        ).fetchone():
+            continue
+        (text,) = conn.execute(
+            "SELECT text FROM prompts WHERE id = ?", (prompt_id,)
+        ).fetchone()
+        found = blockers(conn, prompt_id, unused)
+        audit.append(
+            {
+                "id": prompt_id,
+                "text_excerpt": " ".join(text.split())[:EXCERPT],
+                "text_sha256": sha256(text),
+                "images_before": len(sources[prompt_id]),
+                "blockers": found,
+                "safe": not found,
+            }
+        )
+    return audit
+
+
+def _delete_orphans(conn, plan):
+    chosen = [orphan for orphan in plan["orphans"] if orphan["safe"]]
+    unused = {row[0] for row in conn.execute(ORPHANED_PROMPTS_SQL)}
+    for orphan in chosen:
+        row = conn.execute(
+            "SELECT text FROM prompts WHERE id = ?", (orphan["id"],)
+        ).fetchone()
+        if row is None or sha256(row[0]) != orphan["text_sha256"]:
+            raise Refused(f"prompt {orphan['id']}: gone or text changed")
+        found = blockers(conn, orphan["id"], unused)
+        if found:
+            raise Refused(f"prompt {orphan['id']}: {', '.join(found)}")
+    ids = [orphan["id"] for orphan in chosen]
+    if _delete_ids(conn, "prompts", ids) != len(ids):
+        raise Refused("an orphan prompt was not deleted")
+    return len(ids)
+
+
+def apply_plan(db, plan, read=None, before_write=None):
+    """Write ``plan`` in one transaction; returns ``(result, orphan audit)``.
+
+    ``before_write()`` runs just before ``BEGIN IMMEDIATE`` (the running
+    check). Raises ``Refused`` (and rolls back) on any drift or failed check.
+    """
+    judge = Judge(db, plan["output_dir"], read)
+    relinks = [item for item in plan["rows"] if item["action"] == "relink"]
+    if before_write is not None:
+        before_write()
+    with db._pair_transaction() as conn:
+        _rejudge(conn, judge, plan)
+        targets = _targets(conn, db, relinks)
+        dropped = sum(
+            _relink(conn, item, targets[item["target_hash"]]) for item in relinks
+        )
+        audit = _audit(conn, relinks)
+        deleted = _delete_orphans(conn, plan) if plan.get("delete_orphans") else 0
+    created = sum(1 for item in relinks if item["target_prompt_id"] == "new")
+    result = {
+        "relinked": len(relinks) - dropped,
+        "duplicates_dropped": dropped,
+        "prompts_created": len(
+            {i["target_hash"] for i in relinks if i["target_prompt_id"] == "new"}
+        ),
+        "rows_to_new_prompts": created,
+        "kept": sum(1 for item in plan["rows"] if item["action"] == "keep"),
+        "orphans_deleted": deleted,
+    }
+    return result, audit
+
+
+def counts(path, readonly=True):
+    """Table counts plus the integrity and foreign-key checks.
+
+    Read-only for a library; the tool's own backup is opened read-write so
+    a WAL file leaves no side files behind once it closes.
+    """
+    uri = Path(os.path.abspath(path)).as_uri() + ("?mode=ro" if readonly else "")
+    conn = sqlite3.connect(uri, uri=True)
+    try:
+        found = {
+            table: conn.execute(f'SELECT COUNT(*) FROM "{table}"').fetchone()[0]
+            for table in COUNTED  # nosec B608 - fixed table names
+        }
+        found["integrity_check"] = conn.execute("PRAGMA integrity_check").fetchone()[0]
+        found["foreign_key_check"] = len(
+            conn.execute("PRAGMA foreign_key_check").fetchall()
+        )
+        return found
+    finally:
+        conn.close()
+
+
+def _print_counts(before, after, out):
+    out.write(f"{'':24}{'before':>12}{'after':>12}\n")
+    for key in before:
+        out.write(f"{key:24}{before[key]!s:>12}{after[key]!s:>12}\n")
+
+
+def _summary(plan):
+    actions = [item["action"] for item in plan["rows"]]
+    return "  ".join(
+        f"{name}: {actions.count(name)}" for name in ("relink", "keep", "skip")
+    )
+
+
+def _cell(text, limit=EXCERPT):
+    """One table cell: whitespace folded, cut to ``limit`` (None: whole)."""
+    folded = " ".join(str(text).split())
+    return (folded if limit is None else folded[:limit]).replace("|", "\\|")
+
+
+def _relative(path, root):
+    try:
+        return os.path.relpath(path, root)
+    except ValueError:  # another drive on Windows
+        return path
+
+
+def markdown(plan, prompt_texts):
+    """The maintainer's list of the plan."""
+    root = plan["output_dir"]
+    rel = lambda path: _relative(path, root)  # noqa: E731
+    rows = plan["rows"]
+    lines = [
+        "# #247 repair plan (dry run)",
+        "",
+        f"Created {plan['created']}; output folder `{root}`. {_summary(plan)}.",
+        "",
+        f"## Relink ({sum(i['action'] == 'relink' for i in rows)})",
+        "",
+        "| row | file | linked prompt | linked text | target | target hash "
+        "| target text |",
+        "|---:|---|---:|---|---|---|---|",
+    ]
+    for item in (i for i in rows if i["action"] == "relink"):
+        lines.append(
+            f"| {item['row_id']} | {_cell(rel(item['file']), None)} | "
+            f"{item['current_prompt_id']} | "
+            f"{_cell(prompt_texts.get(item['current_prompt_id'], ''))} | "
+            f"{item['target_prompt_id']} | `{item['target_hash'][:12]}` | "
+            f"{_cell(item['target_text'])} |"
+        )
+    lines += ["", f"## Keep ({sum(i['action'] == 'keep' for i in rows)})", ""]
+    lines += ["| row | file | prompt |", "|---:|---|---:|"]
+    for item in (i for i in rows if i["action"] == "keep"):
+        lines.append(
+            f"| {item['row_id']} | {_cell(rel(item['file']), None)} | "
+            f"{item['current_prompt_id']} |"
+        )
+    lines += ["", f"## Skip ({sum(i['action'] == 'skip' for i in rows)})", ""]
+    lines += ["| row | file | prompt | reason |", "|---:|---|---:|---|"]
+    for item in (i for i in rows if i["action"] == "skip"):
+        reason = item["reason"] + (
+            f" ({item['decision']})" if item.get("decision") else ""
+        )
+        lines.append(
+            f"| {item['row_id']} | {_cell(rel(item['file']), None)} | "
+            f"{item['current_prompt_id']} | {reason} |"
+        )
+    choice = "ON (--delete-orphans)" if plan["delete_orphans"] else "OFF"
+    lines += [
+        "",
+        f"## Prompts that would lose every link ({len(plan['orphans'])})",
+        "",
+        f"Deleting the safe ones: {choice}.",
+        "",
+        "| id | text | images before | blockers | safe to delete |",
+        "|---:|---|---:|---|---|",
+    ]
+    for orphan in plan["orphans"]:
+        lines.append(
+            f"| {orphan['id']} | {_cell(orphan['text_excerpt'])} | "
+            f"{orphan['images_before']} | {', '.join(orphan['blockers']) or 'none'} | "
+            f"{'yes' if orphan['safe'] else 'no'} |"
+        )
+    simulated = plan.get("simulated")
+    if simulated:
+        lines += ["", "## Rehearsed on the dry run's copy", ""]
+        lines += ["| table | before | after |", "|---|---:|---:|"]
+        for key, value in simulated["before"].items():
+            lines.append(f"| {key} | {value} | {simulated['after'][key]} |")
+    return "\n".join(lines) + "\n"
+
+
+def _texts(db):
+    return dict(
+        db.model.get_connection().execute(
+            "SELECT id, text FROM prompts WHERE id IN (SELECT prompt_id "
+            "FROM generated_images)"
+        )
+    )
+
+
+def _rehearse(source, scratch, name, plan):
+    """``apply_plan`` on a fresh backup of ``source``: (counts, result, audit)."""
+    copy = os.path.join(scratch, name)
+    backup_database(source, copy)
+    db = PromptDatabase(copy)
+    try:
+        before = counts(copy)
+        result, audit = apply_plan(db, plan)
+    finally:
+        db.close_all()
+    return {"before": before, "after": counts(copy), "result": result}, audit
+
+
+def dry_run(args, out):
+    with tempfile.TemporaryDirectory() as scratch:
+        copy = os.path.join(scratch, "judge.db")
+        backup_database(args.database, copy)
+        db = PromptDatabase(copy)
+        try:
+            plan = build_plan(db, os.path.abspath(args.output_dir), args.database)
+            texts = _texts(db)
+        finally:
+            db.close_all()
+        simulated, audit = _rehearse(args.database, scratch, "relinks.db", plan)
+        plan.update(orphans=audit, delete_orphans=bool(args.delete_orphans))
+        if plan["delete_orphans"]:
+            # The plan as the apply will write it, deletes included
+            simulated, _audit = _rehearse(args.database, scratch, "deletes.db", plan)
+    plan["simulated"] = simulated
+    plan_path = Path(args.plan)
+    plan_path.write_text(
+        json.dumps(plan, indent=1, ensure_ascii=False) + "\n", encoding="utf-8"
+    )
+    list_path = plan_path.with_name(plan_path.stem + ".list.md")
+    list_path.write_text(markdown(plan, texts), encoding="utf-8")
+    out.write(f"{_summary(plan)}\n")
+    out.write(
+        f"Prompts that would lose every link: {len(audit)} "
+        f"({sum(o['safe'] for o in audit)} safe to delete; deleting "
+        f"{'ON' if plan['delete_orphans'] else 'OFF'})\n"
+    )
+    out.write(f"Plan: {plan_path}\nList: {list_path}\n")
+    out.write("Dry run: nothing was changed.\n")
+    return 0
+
+
+def _is_live(database, comfyui_dir):
+    live = os.path.join(comfyui_dir, *LIVE_LIBRARY)
+    return library_identity(database) == library_identity(live)
+
+
+def _stamp():
+    return datetime.datetime.now().strftime("%Y%m%d_%H%M%S")
+
+
+def schema_of(path):
+    """``sqlite_master`` and ``user_version`` from a read-only connection."""
+    uri = Path(os.path.abspath(path)).as_uri() + "?mode=ro"
+    conn = sqlite3.connect(uri, uri=True)
+    try:
+        master = conn.execute(
+            "SELECT type, name, tbl_name, sql FROM sqlite_master ORDER BY type, name"
+        ).fetchall()
+        return master, conn.execute("PRAGMA user_version").fetchone()[0]
+    finally:
+        conn.close()
+
+
+def running_reasons(args, probe, handles):
+    """Why writing must not start now (empty: go ahead).
+
+    Another process holding the library open always counts, also in a
+    rehearsal; ComfyUI's processes and port count unless rehearsing.
+    """
+    held = [
+        reason
+        for reason in handles(args.database)
+        if not (reason == NO_PROC and args.i_stopped_comfyui)
+    ]
+    return held + ([] if args.rehearsal else probe(args.comfyui_dir, args.port))
+
+
+def _preflight(args, plan, probe, handles):
+    """The refusals before any backup or write, as a message (None: go on)."""
+    if plan.get("version") != PLAN_VERSION:
+        return "not a plan of this tool"
+    if args.rehearsal and _is_live(args.database, args.comfyui_dir):
+        return f"--rehearsal is for a copy; this is the library of {args.comfyui_dir}"
+    if plan.get("source_db") != library_identity(args.database):
+        return f"this plan was made for {plan.get('source_db')}"
+    if not os.path.isfile(os.path.join(args.comfyui_dir, "main.py")):
+        return f"no main.py in {args.comfyui_dir}: not a ComfyUI folder"
+    reasons = running_reasons(args, probe, handles)
+    if reasons:
+        return "ComfyUI may be running: " + "; ".join(reasons)
+    return None
+
+
+def _backup(args, before):
+    """The verified backup's path; raises ``Refused``."""
+    folder = args.backup_dir or os.path.dirname(os.path.abspath(args.database))
+    name = f"{os.path.basename(args.database)}.pre-247-repair_{_stamp()}"
+    backup = os.path.join(folder, name)
+    if os.path.exists(backup):
+        raise Refused(f"the backup {backup} already exists")
+    backup_database(args.database, backup)
+    saved = counts(backup, readonly=False)
+    if saved != before or saved["integrity_check"] != "ok":
+        raise Refused(f"the backup {backup} does not match the library")
+    return backup
+
+
+def _open_unchanged(path):
+    """``PromptDatabase(path)``; raises ``Refused`` when opening it changed
+    its schema (a migration of a newer checkout, committed on open)."""
+    schema = schema_of(path)
+    db = PromptDatabase(path)
+    db.model.get_connection()
+    if schema_of(path) != schema:
+        db.close_all()
+        raise Refused(
+            "opening the library with this checkout changed its schema (that "
+            "change is committed); run the tool from a checkout at the live "
+            "code version"
+        )
+    return db
+
+
+def apply(args, out, probe, handles):
+    with open(args.apply, encoding="utf-8") as stream:
+        plan = json.load(stream)
+    refusal = _preflight(args, plan, probe, handles)
+    if refusal:
+        out.write(f"REFUSED: {refusal}\n")
+        return 2
+    if args.rehearsal:
+        out.write("(rehearsal on a copy: ComfyUI's process and port not checked)\n")
+    before = counts(args.database)
+    try:
+        backup = _backup(args, before)
+    except Refused as error:
+        out.write(f"REFUSED: {error}\n")
+        return 2
+    out.write(f"Backup: {backup}\n")
+
+    def still_stopped():
+        reasons = running_reasons(args, probe, handles)
+        if reasons:
+            raise Refused("ComfyUI may be running: " + "; ".join(reasons))
+
+    try:
+        db = _open_unchanged(args.database)
+    except Refused as error:
+        out.write(f"REFUSED: {error}. Backup: {backup}\n")
+        return 2
+    try:
+        result, _audit = apply_plan(db, plan, before_write=still_stopped)
+    except Exception as error:
+        out.write(
+            f"REFUSED (rolled back, nothing changed): {error}. Backup: {backup}\n"
+        )
+        return 2
+    finally:
+        db.close_all()
+    after = counts(args.database)
+    _print_counts(before, after, out)
+    out.write(f"Applied: {json.dumps(result, sort_keys=True)}\n")
+    if after["integrity_check"] != "ok" or after["foreign_key_check"]:
+        out.write(f"WARNING: integrity or foreign-key check failed; restore {backup}\n")
+        return 1
+    return 0
+
+
+def main(argv=None, out=sys.stdout, probe=comfyui_running, handles=open_by_others):
+    parser = argparse.ArgumentParser(description=__doc__.split("\n", 1)[0])
+    parser.add_argument("database", help="the library database (prompts.db)")
+    parser.add_argument("--output-dir", help="ComfyUI's output folder (dry run)")
+    parser.add_argument(
+        "--plan", default="repair_foreign_links_plan.json", help="plan to write"
+    )
+    parser.add_argument(
+        "--delete-orphans",
+        action="store_true",
+        help="record in the plan that the safe orphan prompts are deleted",
+    )
+    parser.add_argument("--apply", metavar="PLAN", help="write this reviewed plan")
+    parser.add_argument("--comfyui-dir", help="the ComfyUI folder (apply)")
+    parser.add_argument("--backup-dir", help="where the backup goes (apply)")
+    parser.add_argument("--port", type=int, default=COMFY_PORT)
+    parser.add_argument(
+        "--rehearsal",
+        action="store_true",
+        help="apply to a copy while ComfyUI runs (never its own library)",
+    )
+    parser.add_argument(
+        "--i-stopped-comfyui",
+        action="store_true",
+        help="without /proc: say ComfyUI is stopped (the port is still checked)",
+    )
+    args = parser.parse_args(argv)
+    if not os.path.isfile(args.database):
+        parser.error(f"no database at {args.database}")
+    if args.apply:
+        if not args.comfyui_dir:
+            parser.error("--apply needs --comfyui-dir (the running check)")
+        return apply(args, out, probe, handles)
+    if not args.output_dir:
+        parser.error("the dry run needs --output-dir")
+    if Path(args.plan).suffix.lower() == ".md":
+        parser.error("--plan is the JSON plan; the list is written beside it")
+    return dry_run(args, out)
+
+
+if __name__ == "__main__":
+    sys.exit(main())

utils/foreign_runs.py

@@ -0,0 +1,142 @@
+"""The monitor's path for outputs of workflows without a PromptManager node (#247).
+
+A foreign output (no positive PromptManager node in its graph) is decided by
+the ladder the output scan uses (``utils.file_prompt.decide_file_prompt``)
+with the ``MONITOR`` policy: certain files are stored at once, files only
+recorded history can decide wait for their run's outputs to be sealed, and
+everything else stays unlinked. This module holds the parts that do not
+need the monitor's waiting sets: the setting, the decision call and the
+store. The waiting and the re-check stay on ``ImageGenerationHandler``.
+"""
+
+import os
+from dataclasses import dataclass
+from typing import Any, Callable, Mapping, Optional, Tuple
+
+from .file_prompt import (
+    NOT_SAVED,
+    NOT_SAVED_WIRED,
+    PAIR,
+    POSITIVE,
+    Decision,
+    Policy,
+    decide_file_prompt,
+)
+from .hashing import generate_prompt_hash
+from .usage_tracking import survive_switch
+
+try:
+    from ..database.pair_ingestion import RUN_MARKS, ingest_resolved_pair
+    from ..database.run_attribution import recorded_output_lookup
+    from ..database.scanned_prompts import store_scanned_prompt_or_raise
+except ImportError:
+    from database.pair_ingestion import RUN_MARKS, ingest_resolved_pair
+    from database.run_attribution import recorded_output_lookup
+    from database.scanned_prompts import store_scanned_prompt_or_raise
+
+# Decisions that name a prompt to store
+CERTAIN = (PAIR, POSITIVE)
+# Decisions of a run whose prompt has Save to library off, or wired from a
+# node the file cannot show (#90): final, nothing is stored, nothing waits
+NOT_SAVED_KINDS = (NOT_SAVED, NOT_SAVED_WIRED)
+# Sources certain without a recorded run: the graph alone, or its recorded saver
+ANCHORED_SOURCES = ("inspector", "recorded-run")
+
+
+@dataclass(frozen=True)
+class ForeignRecheck:
+    """``schedule`` snapshot: the one re-check of a foreign output after its
+    run's outputs were sealed. ``reference`` is the sealed output reference,
+    the spelling history recorded (``recorded_output_lookup(reference=)``)."""
+
+    reference: Mapping[str, Any]
+
+
+def save_foreign_runs() -> bool:
+    """Whether a foreign output may be stored now, read for every file.
+
+    ``GalleryConfig.SAVE_FOREIGN_RUNS`` and ``MONITORING_ENABLED``: a
+    monitoring switch saved off stops foreign writes at once, although the
+    running observer watches until a restart. True when the configuration
+    cannot be imported (the defaults).
+    """
+    try:
+        from ..py.config import GalleryConfig
+    except ImportError:
+        try:
+            from py.config import GalleryConfig
+        except ImportError:
+            return True
+    return (
+        GalleryConfig.SAVE_FOREIGN_RUNS is not False
+        and GalleryConfig.MONITORING_ENABLED is not False
+    )
+
+
+def history_lookup(
+    db: Any, image_path: str, reference: Optional[Mapping[str, Any]] = None
+) -> Callable[[], Any]:
+    """The ladder's ``lookup`` for one file: its ``RecordedOutput`` or None.
+
+    With ``reference`` the file is spelled as history recorded it. Failures
+    are logged by ``recorded_output_lookup`` and name nothing.
+    """
+    lookup = recorded_output_lookup(
+        db, reference=dict(reference) if reference is not None else None
+    )
+    return lambda: lookup(image_path)
+
+
+def decide_foreign(
+    image_path: str,
+    metadata: Any,
+    *,
+    in_output: bool,
+    lookup: Optional[Callable[[], Any]],
+) -> Decision:
+    """The ``MONITOR`` ladder for one foreign output (the size is read only
+    when the ladder needs it)."""
+
+    def file_size() -> Optional[int]:
+        try:
+            return os.path.getsize(image_path)
+        except OSError:
+            return None
+
+    return decide_file_prompt(
+        metadata,
+        file_name=os.path.basename(image_path),
+        file_size=file_size,
+        in_output=in_output,
+        lookup=lookup,
+        policy=Policy.MONITOR,
+    )
+
+
+def store_foreign(
+    db: Any, image_path: str, decision: Decision
+) -> Tuple[str, Optional[int]]:
+    """Store a certain decision: ``(status, prompt id)``.
+
+    One write through ``survive_switch``, so it runs again on the moved
+    library when a Move library switched while it waited (#249 S6b). Created
+    prompts carry ``RUN_MARKS``. A file linked to another prompt is a
+    ``conflict`` and nothing is written. Database errors are raised.
+    """
+
+    def store():
+        if decision.kind == PAIR:
+            result = ingest_resolved_pair(
+                db, image_path, decision.canonical, decision.resolution, marks=RUN_MARKS
+            )
+            return result["status"], result.get("image_id")
+        return store_scanned_prompt_or_raise(
+            db, image_path, decision.canonical, decision.text, RUN_MARKS
+        )
+
+    status, linked = survive_switch(db, store)
+    if decision.kind != PAIR or linked is None:
+        return status, linked
+    # The pair ingester reports the image row; its prompt is the positive's.
+    prompt = db.get_prompt_by_hash(generate_prompt_hash(decision.text))
+    return status, prompt["id"] if prompt else None

utils/hashing.py

@@ -58,11 +58,7 @@ def generate_prompt_hash(text: str) -> str:
     if not isinstance(text, str):
         raise TypeError("Text must be a string")
 
-    # Normalize the text by stripping whitespace and converting to lowercase
-    # for consistent hashing regardless of minor formatting differences
-    normalized_text = text.strip().lower()
-
-    return hashlib.sha256(normalized_text.encode("utf-8")).hexdigest()
+    return hashlib.sha256(prompt_key(text).encode("utf-8")).hexdigest()
 
 
 def stored_prompt_text(text: str) -> str:
@@ -93,6 +89,34 @@ STRIPPED_CODE_POINTS = (
 _STRIP_CODES = ",".join(map(str, STRIPPED_CODE_POINTS))
 
 
+def prompt_key(text: str) -> str:
+    """The dedup key: the stored text with case folded by ``str.lower()``.
+
+    Two texts with one key are one prompt (the hash is sha256 of the key), so
+    a run belongs to the history of the prompt with its key (#460).
+    """
+    return stored_prompt_text(text).lower()
+
+
+def _sql_key(value):
+    return prompt_key(value) if isinstance(value, str) else None
+
+
+def _sql_hash(value):
+    return generate_prompt_hash(value) if isinstance(value, str) else None
+
+
+def register_prompt_functions(conn) -> None:
+    """Give ``conn`` ``pm_prompt_key(x)`` and ``pm_prompt_hash(x)`` (#460).
+
+    SQLite's lower() folds ASCII only; these call Python, so SQL gets exactly
+    ``prompt_key`` / ``generate_prompt_hash`` (NULL for a non-text value).
+    Never used in schema objects, so the file stays readable without them.
+    """
+    conn.create_function("pm_prompt_key", 1, _sql_key, deterministic=True)
+    conn.create_function("pm_prompt_hash", 1, _sql_hash, deterministic=True)
+
+
 def stored_prompt_text_sql(expression: str) -> str:
     """``stored_prompt_text`` as an SQLite expression over ``expression``.
 

utils/image_monitor.py

@@ -33,6 +33,17 @@ from typing import Callable, Optional, Dict, Any, List, Tuple
 from watchdog.observers import Observer
 from watchdog.events import FileSystemEventHandler
 
+from .file_prompt import HISTORY, NOT_SAVED, NOT_SAVED_WIRED
+from .foreign_runs import (
+    ANCHORED_SOURCES,
+    CERTAIN,
+    NOT_SAVED_KINDS,
+    ForeignRecheck,
+    decide_foreign,
+    history_lookup,
+    save_foreign_runs,
+    store_foreign,
+)
 from .hashing import generate_prompt_hash
 from .metadata_extractor import ComfyUIMetadataExtractor
 from .parallel import note_refusal, refused_at_shutdown
@@ -153,6 +164,11 @@ class ImageGenerationHandler(FileSystemEventHandler):
         # Sealed re-checks ``schedule`` refused (the path still pending, or the
         # pending set full): retried each time a pending image finishes.
         self._history_due: "OrderedDict[str, None]" = OrderedDict()
+        # The same two sets for foreign outputs (#247), apart so a large
+        # foreign run never evicts a waiting keyless file; the due set keeps
+        # each file's sealed output reference.
+        self._awaiting_foreign: "OrderedDict[str, Tuple[str, float]]" = OrderedDict()
+        self._foreign_due: "OrderedDict[str, dict]" = OrderedDict()
         self._awaiting_lock = threading.Lock()
         self._sealed_subscribed = False
         # Set by ``close``: no new work, no new worker, ever (#302).
@@ -253,6 +269,8 @@ class ImageGenerationHandler(FileSystemEventHandler):
         with self._awaiting_lock:
             self._awaiting.clear()
             self._history_due.clear()
+            self._awaiting_foreign.clear()
+            self._foreign_due.clear()
         with self._pending_lock:
             self._pending.clear()
             worker, self._worker = self._worker, None
@@ -438,6 +456,9 @@ class ImageGenerationHandler(FileSystemEventHandler):
             if prompt_snapshot is HISTORY_ONLY:
                 self._recheck_history(image_path)
                 return
+            if isinstance(prompt_snapshot, ForeignRecheck):
+                self._recheck_foreign(image_path, prompt_snapshot.reference)
+                return
             self.logger.info(f"Processing image: {image_path}")
 
             if not os.path.exists(image_path):
@@ -457,17 +478,12 @@ class ImageGenerationHandler(FileSystemEventHandler):
             # negative-prompt PromptManager nodes in the workflow.
             state, current_prompt = self._find_prompt_from_metadata(metadata)
             if state == "opted_out":
-                self.logger.info(
-                    "Save to library is off for this run's prompt, skipping image: "
-                    f"{image_path}"
-                )
+                self._log_not_saved(image_path)
                 return
             if state == "foreign":
-                # Another workflow saved this image: it belongs to no prompt of
-                # ours, and must not consume a batch queue entry.
-                self.logger.info(
-                    f"No PromptManager prompt in workflow, skipping image: {image_path}"
-                )
+                # Another workflow saved this image: never a batch queue entry
+                # of ours; its own graph may name its prompt (#247).
+                self._ingest_foreign(image_path, metadata)
                 return
             if current_prompt:
                 self.logger.info(
@@ -589,36 +605,55 @@ class ImageGenerationHandler(FileSystemEventHandler):
             resolved = os.path.abspath(image_path)
         return os.path.normcase(resolved)
 
-    def _prune_awaiting(self) -> None:
-        """Drop expired and surplus entries (``_awaiting_lock`` held)."""
+    def _prune_awaiting(self, store=None) -> List[Tuple[str, bool]]:
+        """Drop expired and surplus entries of ``store`` (the keyless set by
+        default; ``_awaiting_lock`` held). Returns the dropped foreign files
+        as ``(path, evicted)`` for ``_log_dropped`` after the lock."""
+        store = self._awaiting if store is None else store
         now = time.monotonic()
-        for key in [k for k, (_p, due) in self._awaiting.items() if due <= now]:
-            del self._awaiting[key]
-        while len(self._awaiting) > MAX_AWAITING_HISTORY:
-            self._awaiting.popitem(last=False)
+        expired = [k for k, (_p, due) in store.items() if due <= now]
+        dropped = [(store.pop(key)[0], False) for key in expired]
+        while len(store) > MAX_AWAITING_HISTORY:
+            dropped.append((store.popitem(last=False)[1][0], True))
+        return dropped if store is self._awaiting_foreign else []
 
-    def _await_history(self, image_path: str) -> None:
+    def _log_dropped(self, dropped: List[Tuple[str, bool]]) -> None:
+        """Log foreign files that stopped waiting (outside ``_awaiting_lock``)."""
+        for image_path, evicted in dropped:
+            name = os.path.basename(image_path)
+            if evicted:
+                self.logger.info(
+                    f"Too many files wait for their run; {name} left unlinked"
+                )
+            else:
+                self.logger.info(f"No recorded run for {name}; left unlinked")
+
+    def _await_history(self, image_path: str, store=None) -> None:
         """Re-check ``image_path`` once, history only, when its run is sealed.
 
-        A closed handler re-checks nothing.
+        ``store`` is the waiting set (``_awaiting_foreign`` for a foreign
+        output). A closed handler re-checks nothing.
         """
         if self._closed:
             return
         if not self._sealed_subscribed:
             subscribe_outputs_sealed(self._on_outputs_sealed)
             self._sealed_subscribed = True
+        store = self._awaiting if store is None else store
         key = self._awaiting_key(image_path)
         with self._awaiting_lock:
-            self._awaiting.pop(key, None)
-            self._awaiting[key] = (
-                image_path,
-                time.monotonic() + AWAIT_HISTORY_SECONDS,
-            )
-            self._prune_awaiting()
+            store.pop(key, None)
+            store[key] = (image_path, time.monotonic() + AWAIT_HISTORY_SECONDS)
+            dropped = self._prune_awaiting(store)
+        self._log_dropped(dropped)
 
-    def _stop_awaiting(self, image_path: str) -> None:
+    def _await_foreign(self, image_path: str) -> None:
+        self._await_history(image_path, self._awaiting_foreign)
+
+    def _stop_awaiting(self, image_path: str, store=None) -> None:
+        store = self._awaiting if store is None else store
         with self._awaiting_lock:
-            self._awaiting.pop(self._awaiting_key(image_path), None)
+            store.pop(self._awaiting_key(image_path), None)
 
     def awaiting_history(self) -> List[str]:
         """Normalized paths still waiting for their run's recorded outputs."""
@@ -626,6 +661,14 @@ class ImageGenerationHandler(FileSystemEventHandler):
             self._prune_awaiting()
             return list(self._awaiting)
 
+    def awaiting_foreign(self) -> List[str]:
+        """Normalized paths of foreign outputs waiting for their run (#247)."""
+        with self._awaiting_lock:
+            dropped = self._prune_awaiting(self._awaiting_foreign)
+            waiting = list(self._awaiting_foreign)
+        self._log_dropped(dropped)
+        return waiting
+
     def _on_outputs_sealed(self, run_id: str, references: List[dict]) -> None:
         """A run's outputs were recorded: re-check its waiting images once.
 
@@ -637,6 +680,7 @@ class ImageGenerationHandler(FileSystemEventHandler):
         due = []
         with self._awaiting_lock:
             self._prune_awaiting()
+            dropped = self._prune_awaiting(self._awaiting_foreign)
             for reference in references:
                 if reference.get("folder_type") != "output":
                     continue
@@ -647,39 +691,53 @@ class ImageGenerationHandler(FileSystemEventHandler):
                 )
                 waiting = self._awaiting.pop(key, None)
                 if waiting is not None:
-                    due.append(waiting[0])
-        for image_path in due:
-            self._schedule_recheck(image_path)
+                    due.append((waiting[0], None))
+                waiting = self._awaiting_foreign.pop(key, None)
+                if waiting is not None:
+                    due.append((waiting[0], dict(reference)))
+        self._log_dropped(dropped)
+        for image_path, reference in due:
+            self._schedule_recheck(image_path, reference)
 
-    def _schedule_recheck(self, image_path: str) -> None:
-        """Schedule the history-only re-check, or keep it due for later.
+    @staticmethod
+    def _recheck_snapshot(reference):
+        return HISTORY_ONLY if reference is None else ForeignRecheck(reference)
+
+    def _schedule_recheck(self, image_path: str, reference=None) -> None:
+        """Schedule the history-only re-check (with the sealed ``reference``:
+        a foreign output's), or keep it due for later.
 
         A closed handler drops it.
         """
         if self._closed:
             return
-        if self.schedule(image_path, HISTORY_ONLY):
+        if self.schedule(image_path, self._recheck_snapshot(reference)):
             return
         with self._awaiting_lock:
             # ``close`` sets the flag before ``shutdown`` clears this set, so
             # a close that raced the refusal above is seen here.
             if self._closed:
                 return
-            self._history_due.pop(image_path, None)
-            self._history_due[image_path] = None
-            while len(self._history_due) > MAX_AWAITING_HISTORY:
-                self._history_due.popitem(last=False)
+            due = self._history_due if reference is None else self._foreign_due
+            due.pop(image_path, None)
+            due[image_path] = reference
+            evicted = []
+            while len(due) > MAX_AWAITING_HISTORY:
+                evicted.append(due.popitem(last=False)[0])
+        if reference is not None:
+            self._log_dropped([(path, True) for path in evicted])
 
     def _drain_history_due(self) -> None:
         """Schedule due re-checks now that a pending slot or path is free."""
         with self._awaiting_lock:
-            if not self._history_due:
-                return
-            due = list(self._history_due)
-        for image_path in due:
-            if self.schedule(image_path, HISTORY_ONLY):
+            due = [(self._history_due, p, None) for p in self._history_due]
+            due += [(self._foreign_due, *item) for item in self._foreign_due.items()]
+        for store, image_path, reference in due:
+            if self.schedule(image_path, self._recheck_snapshot(reference)):
                 with self._awaiting_lock:
-                    self._history_due.pop(image_path, None)
+                    # A newer seal may have replaced it meanwhile: keep that one
+                    if image_path in store and store[image_path] is reference:
+                        del store[image_path]
 
     def _recheck_history(self, image_path: str) -> None:
         """The one history-only re-check of a keyless image (#342)."""
@@ -703,6 +761,108 @@ class ImageGenerationHandler(FileSystemEventHandler):
             metadata or {"file_info": self.get_basic_file_info(image_path)},
         )
 
+    def _ingest_foreign(self, image_path: str, metadata) -> None:
+        """A foreign output at settle: store its certain prompt, or wait for
+        its run's recorded outputs when only history can decide (#247)."""
+        name = os.path.basename(image_path)
+        if not save_foreign_runs():
+            self.logger.info(
+                f"No PromptManager prompt in workflow, skipping image: {image_path}"
+            )
+            return
+        in_output = in_output_folder(image_path, host_output_directory())
+        decision = decide_foreign(
+            image_path, metadata, in_output=in_output, lookup=None
+        )
+        if decision.kind == HISTORY:
+            decision = self._settle_lookup(image_path, metadata)
+            if decision is None:
+                self.logger.debug(f"{name} waits for its run to end")
+                return
+            self._stop_awaiting(image_path, self._awaiting_foreign)
+        if decision.kind in NOT_SAVED_KINDS:
+            self._log_not_saved(image_path, decision.kind)
+        elif decision.kind in CERTAIN:
+            self._store_foreign(image_path, decision)
+        else:
+            self.logger.info(f"No certain prompt for {name}; left unlinked")
+
+    def _log_not_saved(self, image_path: str, kind: str = NOT_SAVED) -> None:
+        """The log line of a file whose run's prompt does not save (#90)."""
+        if kind == NOT_SAVED_WIRED:
+            reason = "is wired from a node the image cannot show"
+        else:
+            reason = "is off"
+        self.logger.info(
+            f"Save to library {reason} for this run's prompt, skipping image: "
+            f"{image_path}"
+        )
+
+    def _settle_lookup(self, image_path: str, metadata):
+        """The settle-time history lookup of a file only history can decide.
+
+        Registered first, so a run sealed before this lookup is not missed.
+        Returns None to keep waiting: no record yet, and the answer is not
+        the inspector's or the recorded saver's (the legacy floor waits for
+        the re-check). With a record, any decision ends the wait. A
+        ``not_saved`` here is the undecided step's (an early one never asks
+        history, so it never reaches this lookup): a recorded saver may still
+        resolve another branch, so without a record it waits too (#90).
+        """
+        self._await_foreign(image_path)
+        found = []
+        lookup = history_lookup(self.db_manager, image_path)
+
+        def recorded():
+            found.append(lookup())
+            return found[-1]
+
+        decision = decide_foreign(image_path, metadata, in_output=True, lookup=recorded)
+        if any(record is not None for record in found):
+            return decision
+        if decision.kind in CERTAIN and decision.source in ANCHORED_SOURCES:
+            return decision
+        return None
+
+    def _recheck_foreign(self, image_path: str, reference) -> None:
+        """The one re-check of a foreign output after its run ended (#247)."""
+        name = os.path.basename(image_path)
+        if not os.path.exists(image_path):
+            return
+        if not save_foreign_runs():
+            self.logger.info(f"Saving prompts from other workflows is off: {name}")
+            return
+        try:
+            metadata = self.metadata_extractor.extract_metadata(image_path)
+        except Exception as error:
+            self.logger.warning(f"Metadata of {name} could not be read: {error}")
+            return
+        lookup = history_lookup(self.db_manager, image_path, reference)
+        decision = decide_foreign(image_path, metadata, in_output=True, lookup=lookup)
+        if decision.kind in NOT_SAVED_KINDS:
+            self._log_not_saved(image_path, decision.kind)
+            return
+        if decision.kind in CERTAIN:
+            self._store_foreign(image_path, decision)
+            return
+        self.logger.info(
+            f"No certain prompt for {name} after its run ended; left unlinked"
+        )
+
+    def _store_foreign(self, image_path: str, decision) -> None:
+        name = os.path.basename(image_path)
+        if not save_foreign_runs():  # switched off while this file was decided
+            self.logger.info(f"Saving prompts from other workflows is off: {name}")
+            return
+        try:
+            status, prompt_id = store_foreign(self.db_manager, image_path, decision)
+        except Exception as error:
+            self.logger.error(f"Failed to store the prompt of {name}: {error}")
+            return
+        self.logger.info(
+            f"Foreign run: {status} prompt {prompt_id} for {name} ({decision.source})"
+        )
+
     def _history_prompt(self, image_path: str) -> Optional[int]:
         """The stored prompt recorded history attributes to this file, if one."""
         try:

utils/inspection_response.py

@@ -548,8 +548,12 @@ def encode_history_inspection_response(
     resolution: dict,
     inspection: "InspectionRead | None",
     library: dict,
+    image_tags: dict | None = None,
 ) -> bytes:
-    """Use occurrence identity with the same current-file evidence projection."""
+    """Use occurrence identity with the same current-file evidence projection.
+
+    *image_tags*, when given, adds the validated ``image_tags`` facet (#457).
+    """
     return _encode_inspection(
         identity={
             "history_version": 1,
@@ -561,4 +565,5 @@ def encode_history_inspection_response(
         resolution=resolution,
         inspection=inspection,
         library=library,
+        image_tags=image_tags,
     )

web/js/gallery-settings.js

@@ -17,6 +17,7 @@
     // The workspace Images defaults; the legacy gallery's defaultViewMode has no masonry.
     const WORKSPACE_LAYOUTS = ["grid", "list", "masonry"];
     const WORKSPACE_DENSITIES = ["comfortable", "large", "compact"];
+    const VIEWER_OPEN_MODES = ["centered", "fullscreen"];
     const LIMIT_OPTIONS = [50, 100, 200, 500];
     const MIN_COLUMNS = 2;
     const MAX_COLUMNS = 12;
@@ -43,6 +44,8 @@
         // #451: Images opens in Masonry unless this browser chose another layout.
         workspaceLayout: "masonry",
         workspaceDensity: "comfortable",
+        // #450 (D156): how the workspace viewer opens.
+        viewerOpenMode: "centered",
     });
 
     function toInt(value) {
@@ -83,6 +86,7 @@
             videoLoop: bool(merged.videoLoop, DEFAULT_SETTINGS.videoLoop),
             workspaceLayout: WORKSPACE_LAYOUTS.includes(merged.workspaceLayout) ? merged.workspaceLayout : DEFAULT_SETTINGS.workspaceLayout,
             workspaceDensity: WORKSPACE_DENSITIES.includes(merged.workspaceDensity) ? merged.workspaceDensity : DEFAULT_SETTINGS.workspaceDensity,
+            viewerOpenMode: VIEWER_OPEN_MODES.includes(merged.viewerOpenMode) ? merged.viewerOpenMode : DEFAULT_SETTINGS.viewerOpenMode,
         };
     }
 
@@ -92,7 +96,7 @@
         return `repeat(${n}, minmax(0, 1fr))`;
     }
 
-    const api = { DEFAULT_SETTINGS, LIMIT_OPTIONS, VIEW_MODES, WORKSPACE_LAYOUTS, WORKSPACE_DENSITIES, normalizeGallerySettings, gridColumnsStyle };
+    const api = { DEFAULT_SETTINGS, LIMIT_OPTIONS, VIEW_MODES, WORKSPACE_LAYOUTS, WORKSPACE_DENSITIES, VIEWER_OPEN_MODES, normalizeGallerySettings, gridColumnsStyle };
     if (typeof module !== "undefined" && module.exports) module.exports = api;
     else root.GallerySettings = api;
 })(typeof window !== "undefined" ? window : globalThis);

web/js/workspace/icons.mjs

@@ -21,7 +21,10 @@ const paths = Object.freeze({
     "tagImage": "M3 4h8l9 9-7 7L3 10V4ZM7 8h.01M18 2v5M15.5 4.5h5",
     "folder": "M3 5h7l2 3h9v12H3z",
     "history": "M3 11a9 9 0 1 1 2 7M3 4v7h7M12 7v6l4 2",
-    "review": "M4 3h16v18H4zM8 7h8M8 12h8M8 17h5"
+    "review": "M4 3h16v18H4zM8 7h8M8 12h8M8 17h5",
+    // #450 (D156): the viewer's Full screen / Centered switch.
+    "maximize": "M4 9V4h5M20 9V4h-5M4 15v5h5M20 15v5h-5",
+    "minimize": "M9 4v5H4M15 4v5h5M9 20v-5H4M15 20v-5h5"
 });
 
 /** Create an approved decorative outline icon without injecting markup. */

web/js/workspace/inspect.mjs

@@ -1,6 +1,6 @@
 // The concept's "Inspect a file" page (docs/concepts/workspace/app.js renderMetadata 740–748, selection 1715–1729):
-// one local file, chosen or dropped, previews inline; the heading's one action opens the shared viewer with the full
-// inspector on a record read entirely in this browser. Nothing is uploaded, imported or stored. The selection, its
+// one local file, chosen or dropped, previews inline and (D155, #448) opens the shared viewer with its details showing
+// at once; the heading's one action opens it again. The record is read entirely in this browser. Nothing is uploaded, imported or stored. The selection, its
 // blob: URL and its record belong to one local session (local-session.mjs) that ends with this page.
 import { createButton, createScope } from '../shared/toolkit/core.js';
 import { pageHeading } from './heading.mjs';
@@ -17,7 +17,8 @@ export const INSPECT_COPY = Object.freeze({
     heading: 'Inspect without importing',
     intro: 'Read embedded metadata in this browser. The file is not uploaded or added to your library.',
     chooser: 'Choose an image or video for a local preview',
-    note: 'Your file stays in this browser. Its metadata opens with “Open file metadata”.',
+    // D155 (#448): choosing a file opens its metadata (D61's note named the heading action).
+    note: 'Your file stays in this browser. Choosing it opens its metadata.',
     toast: 'Local preview only. No file uploaded; metadata is available above.',
 });
 // D62: the concept's image list plus the videos the local readers support (AVI/WMV preview only).
@@ -38,8 +39,11 @@ function element(tag, className, text) {
     return node;
 }
 
-// The status shown for the selection; AVI/WMV's note and an empty status keep the toast's "metadata is available".
-const TOASTABLE = new Set(['', LOCAL_SESSION_MESSAGES.containerOnly]);
+// D155 (#448): a selection with no status, or one without a preview URL (its copy points at the viewer), opens the viewer
+// at once. AVI/WMV's note keeps the toast instead (the viewer could not play it and reads no metadata), and an
+// unreadable file stays on the page with its alert; the heading action opens either.
+const AUTO_OPEN = new Set(['', LOCAL_SESSION_MESSAGES.previewFailed]);
+const TOASTABLE = new Set([LOCAL_SESSION_MESSAGES.containerOnly]);
 // How long a replaced video still starting may stay in place before it is released anyway.
 const STARTING_VIDEO_MS = 15000;
 
@@ -131,9 +135,10 @@ export function mountInspect(main, { parentScope, api, sender }) {
     };
     const session = createLocalSession({ parentScope: sessionScope, pageTarget: null, revokeObjectURL });
     const viewer = createMediaViewer({ api, container: main, parentScope: scope, inspectionController: session.controller, sender,
-        adapter: localFileAdapter, onRestoreFocus: () => { if (open.isConnected) open.focus({ preventScroll: true }); } });
+        adapter: localFileAdapter, onRestoreFocus: () => { if (open.isConnected) open.focus({ preventScroll: true }); },
+        onClose: () => { viewerKey = null; } });
 
-    let media = null, mediaUrl = null, statusLine = null, rejectionLine = null, rejectionTimer = 0, announced = null, chosen = null;
+    let viewerKey = null, media = null, mediaUrl = null, statusLine = null, rejectionLine = null, rejectionTimer = 0, announced = null, chosen = null;
     function statusElement(message) {
         // /* D63 */ one status line for the states the mock never shows.
         const line = element('p', 'small spaced-16', message);
@@ -202,14 +207,25 @@ export function mountInspect(main, { parentScope, api, sender }) {
         else preview.prepend(node);
     }
     scope.cleanup(session.subscribe(state => {
+        // D155: a new selection can start while the viewer shows the previous one; it closes first, so no element holds
+        // the old URL when the session revokes it (the new file's viewer opens once it is ready).
+        if (viewerKey !== null && (state.phase !== 'ready' || session.entry?.key !== viewerKey)) { viewer.close(); viewerKey = null; }
         showMedia(state);
         showStatus(state.message);
         const entry = session.entry;
         setAvailable(state.phase === 'ready' && !!entry);
         if (state.phase === 'ready' && entry && announced !== entry.key) {
             announced = entry.key;
+            // The session selects the record right after it publishes `ready`, so the viewer opens after that, and only
+            // while this selection is still the current one and no other dialog is open (the viewer is replaced). D62:
+            // a video opens it once the inline preview has settled, so a replacement while it starts never tears down a
+            // second starting video in the viewer (that deadlocked WebKit).
+            if (AUTO_OPEN.has(state.message)) whenSettled(media, () => {
+                if (scope.signal.aborted || session.entry !== entry || session.state.phase !== 'ready' || document.querySelector('dialog[open]:not(#viewer-dialog)')) return;
+                openViewer(input);
+            });
             // An unreadable or failed preview says so in its alert; the toast never claims metadata is available.
-            if (TOASTABLE.has(state.message)) showWorkspaceToast(INSPECT_COPY.toast);
+            else if (TOASTABLE.has(state.message)) showWorkspaceToast(INSPECT_COPY.toast);
         }
     }));
     // A replacement or page leave revokes the URL after this subscriber dropped the media; a dispose closes the
@@ -261,10 +277,25 @@ export function mountInspect(main, { parentScope, api, sender }) {
     });
     // A drop that misses the area never navigates away from the page while it is mounted.
     for (const type of ['dragover', 'drop']) scope.listen(window, type, event => event.preventDefault());
-    scope.listen(open, 'click', () => {
+    /** Runs `run` after this task, or once `preview`'s video stops starting (metadata, an error, or the D62 limit). */
+    function whenSettled(preview, run) {
+        if (!preview || !isStarting(preview.node)) { queueMicrotask(run); return; }
+        // `run` checks that the page and the selection are still current, so the page's end may call `done` too.
+        const settled = new AbortController();
+        const done = () => { settled.abort(); clearTimeout(timer); run(); };
+        const timer = setTimeout(done, STARTING_VIDEO_MS);
+        for (const type of ['loadedmetadata', 'error']) preview.node.addEventListener(type, done, { once: true, signal: settled.signal });
+        scope.cleanup(done);
+    }
+    /** The viewer on the current selection, details showing (D155 amends D97); focus returns to `trigger` on close. */
+    function openViewer(trigger) {
         const entry = session.entry;
-        if (open.getAttribute('aria-disabled') === 'true' || !entry || session.state.phase !== 'ready') return;
-        viewer.open({ items: [entry], selectedKey: entry.key, trigger: open });
+        if (!entry || session.state.phase !== 'ready') return;
+        viewer.open({ items: [entry], selectedKey: entry.key, trigger, details: 'shown' });
+        viewerKey = entry.key;
+    }
+    scope.listen(open, 'click', () => {
+        if (open.getAttribute('aria-disabled') !== 'true') openViewer(open);
     });
     /** Settles when no video on the page is still starting; `null` when none is. */
     function leaveGuard() {

web/js/workspace/library-state.mjs

@@ -12,10 +12,22 @@ function parameters(filters,limit) {
     for(const tag of filters.tags || []) if(tag.trim()) query.append('tag',tag.trim());
     const sort=filters.sort || 'newest';
     if(!['newest','oldest','name'].includes(sort)) throw new TypeError('Invalid library sort');
+    // #444: how Tags match (every tag, any of them, or parents no tag's Browse returns, which never goes with tags) and
+    // the output folder below a root.
+    const tagMode=filters.tagMode ?? 'all',folder=filters.folder ?? '';
+    if(!LIBRARY_TAG_MODES.includes(tagMode) || (tagMode==='untagged' && query.has('tag'))) throw new TypeError('Invalid library tag mode');
+    if(typeof folder!=='string' || folder.length>1024) throw new TypeError('Invalid library folder');
+    if(tagMode!=='all') query.set('tag_mode',tagMode);
+    if(folder) query.set('folder',folder);
     query.set('sort',sort);return query;
 }
+/** The Library's tag match values (#444); the first is the default, sent as no parameter. */
+export const LIBRARY_TAG_MODES=Object.freeze(['all','any','untagged']);
 // `variant_total` is the filtered parents' active negative variants (the concept caption's "M negative variants").
-function validatePage(page,cursor) {
+function validatePage(page,cursor,query) {
+    // A tag match or folder the server did not echo was not applied (an older server): never show the whole shelf as filtered.
+    if(query?.has('tag_mode') && page?.tag_mode!==query.get('tag_mode')) throw new TypeError('Unconfirmed library tag mode');
+    if(query?.has('folder') && page?.folder!==query.get('folder')) throw new TypeError('Unconfirmed library folder');
     if(!page || !Array.isArray(page.items) || page.items.some(item=>!parent(item)) || !Number.isSafeInteger(page.total) || page.total<0 || !Number.isSafeInteger(page.variant_total) || page.variant_total<0 || typeof page.has_more!=='boolean' || !(page.next_cursor===null || (typeof page.next_cursor==='string' && page.next_cursor.length)) || page.has_more!==(page.next_cursor!==null) || (page.has_more && cursor===page.next_cursor)) throw new TypeError('Invalid library page');
 }
 export function createLibraryStream({api,onChange=()=>{},restore}) {
@@ -37,7 +49,7 @@ export function createLibraryStream({api,onChange=()=>{},restore}) {
         try {
             const page=await api.get(`/prompt_manager/library?${params}`,{signal:current.signal});
             if(disposed || current!==request) return 0;
-            validatePage(page,cursor);
+            validatePage(page,cursor,query);
             // #437: the storage epoch the shelf's items were read under, which a bulk write sends as its fence; a later
             // page from a restored storage is refused (its items would belong to another database).
             const epoch=typeof page.library_epoch==='string'?page.library_epoch:null;
@@ -113,3 +125,28 @@ export function libraryWindow(count,columns,top,height,focused=-1) {
     if(focused>=0 && focused<count) keep.add(Math.floor(focused/columns));
     return [...keep].sort((a,b)=>a-b).flatMap(row=>Array.from({length:columns},(_,column)=>row*columns+column).filter(index=>index<count));
 }
+
+const byFolderName=(a,b)=>{const x=a.toLowerCase(),y=b.toLowerCase();return x<y?-1:x>y?1:a<b?-1:a>b?1:0;};
+/**
+ * Refine › Folder options (#444 D150): the concept's "All output folders" first, then the folders as a tree built by
+ * path segment, so each folder sits directly under its own parent (never under a longer sibling such as `kiko-v6`) and
+ * folders differing only in case stay distinct siblings. Siblings sort case-insensitively, then by exact spelling. A
+ * label is the last segment indented by depth (the legacy admin select). The chosen folder is kept even when the list
+ * lacks it, with its parents.
+ */
+export function libraryFolderOptions(folders,current='') {
+    const root=new Map();
+    for(const folder of current?[...folders,current]:folders) {
+        let level=root;
+        for(const segment of folder.split('/')) {if(!level.has(segment)) level.set(segment,new Map());level=level.get(segment);}
+    }
+    const options=[['','All output folders']];
+    (function walk(level,prefix,depth) {
+        for(const segment of [...level.keys()].sort(byFolderName)) {
+            const path=prefix?`${prefix}/${segment}`:segment;
+            options.push([path,`${'\u00a0\u00a0'.repeat(depth)}${segment}`]);
+            walk(level.get(segment),path,depth+1);
+        }
+    })(root,'',0);
+    return options;
+}

web/js/workspace/library.mjs

@@ -2,7 +2,7 @@ import {createScope,createButton} from '../shared/toolkit/core.js';
 import {createPromptSender} from './prompt-sender.mjs';
 import {mountPromptBlocks} from './prompt-inspector.mjs';
 import {mountImages} from './images.mjs';
-import {createLibraryStream,createLibraryDetail,libraryPromptRoles} from './library-state.mjs';
+import {createLibraryStream,createLibraryDetail,libraryPromptRoles,LIBRARY_TAG_MODES,libraryFolderOptions} from './library-state.mjs';
 import {promptRoles} from './prompt-blocks.mjs';
 import {createLibraryEditor,showWorkspaceToast} from './library-editor.mjs';
 import {createLibraryEditorController} from './library-editor-state.mjs';
@@ -31,6 +31,8 @@ const metadataDrafts=new Map();
 const archivedVisible=new Map();
 // The concept's sorts first, then the Library's appended Oldest first (D14); D153 (#447) drops "Highest rated".
 const LIBRARY_SORTS=Object.freeze([['newest','Recently added'],['name','Name A–Z'],['oldest','Oldest first']]);
+// Refine › Tag match (#444 D149, D150), in LIBRARY_TAG_MODES order.
+const TAG_MATCH_OPTIONS=Object.freeze([['all','All of these tags'],['any','Any of these tags'],['untagged','Prompts without tags']]);
 // A card's focusable controls in order: the D148 selection checkbox, then its buttons (#437).
 const CARD_CONTROLS='button, input[type="checkbox"]';
 const CATEGORIES_FAILED='Categories could not be loaded. Open the Prompt library again to retry.';
@@ -83,9 +85,9 @@ export function onLibraryWrite(listener) {
  * with the existing Library filters, and the heading takes focus as after a save.
  */
 export function browseLibrary(filters) {
-    const {text='',category='',tags=[],sort='newest'}=filters || {};
-    if(typeof text!=='string' || typeof category!=='string' || !Array.isArray(tags) || tags.some(tag=>typeof tag!=='string') || typeof sort!=='string') throw new TypeError('Invalid Library filters');
-    shelfVisit={filters:{text,category,tags:[...tags],sort},anchor:null};
+    const {text='',category='',tags=[],sort='newest',tagMode='all',folder=''}=filters || {};
+    if(typeof text!=='string' || typeof category!=='string' || !Array.isArray(tags) || tags.some(tag=>typeof tag!=='string') || typeof sort!=='string' || !LIBRARY_TAG_MODES.includes(tagMode) || typeof folder!=='string') throw new TypeError('Invalid Library filters');
+    shelfVisit={filters:{text,category,tags:[...tags],sort,tagMode,folder},anchor:null};
     focusHeadingOnMount=true;window.location.hash='#/library';
 }
 /** Remember the selection a reconciled save names, so the destination history opens on it once. */
@@ -158,15 +160,16 @@ export function mountLibrary(main,api,parentScope,{positiveId,savedState=null,on
         const visit=saved?null:shelfVisit, mountedGeneration=writeGeneration;
         // An invalidated visit keeps only filters and anchor: the stream restarts and heights are measured again.
         const restorable=visit?.state?visit:null;
-        const filters={text:'',category:'',tags:[],sort:'newest',...visit?.filters,...saved};
+        const filters={text:'',category:'',tags:[],sort:'newest',tagMode:'all',folder:'',...visit?.filters,...saved};
         // D153: a sort the shelf no longer offers (an older "Highest rated" visit) reads as Recently added.
         if(!LIBRARY_SORTS.some(([value])=>value===filters.sort)) filters.sort='newest';
         const section=el('section',undefined,'library-shelf');section.setAttribute('aria-label','Prompt library');
         // The shared concept toolbar (app.js toolbar(), a347–360): Tags live in "Refine this view" (D14), no Refresh (D15).
         // Its own scope, disposed after the visit is saved: removing it first would shorten the page and clamp scrollY.
         const toolbarScope=createScope();
-        // D44: Filters reads "Filters · 1" while a Tags filter (from Refine or a Browse) narrows the shelf.
-        const refinements=()=>filters.tags.length?1:0;
+        // D44: Filters reads "Filters · N" while Refine filters (from Refine or a Browse) narrow the shelf: Tags, a Folder
+        // and a Tag match other than All (#444 D149, D150) each count once.
+        const refinements=()=>(filters.tags.length?1:0)+(filters.folder?1:0)+(filters.tagMode!=='all'?1:0);
         const toolbar=mountToolbar(section,{scope:toolbarScope,categories:filters.category?[filters.category]:[],sorts:LIBRARY_SORTS,
             value:{q:filters.text,category:filters.category,sort:filters.sort},activeFilters:refinements(),onChange:applyToolbar,onFilters:openFilters,
             onSaveView:trigger=>{const value=toolbar.value();onSaveView(trigger,libraryViewState({...filters,text:value.q.trim(),category:value.category,sort:value.sort}));}});
@@ -213,7 +216,7 @@ export function mountLibrary(main,api,parentScope,{positiveId,savedState=null,on
         // Nothing follows the grid in the concept; the legacy Library stays one click away on Overview (spec §5).
         continuation.append(message,retry,more);section.append(optionsStatus,exportStatus,selection.element,caption,grid,continuation);main.append(section);
         const cards=new Map(), rowNodes=new Map(), spacers=new Map(), heights=new Map(), pendingMeasurements=new Map();
-        let state,frame=0,loadFrame=0,observer,optionsRequest,focusOverride=-1;
+        let state,frame=0,loadFrame=0,observer,optionsRequest,folderRequest,focusOverride=-1;
         const columns=()=>window.innerWidth>=1700?3:window.innerWidth<=700?1:2;
         const readInlineSize=()=>Math.round(grid.getBoundingClientRect().width*64)/64;
         let inlineSize=readInlineSize(), columnCount=columns(), layout={rows:[],height:0}, dirty=true, items, indexesById=new Map();
@@ -394,7 +397,10 @@ export function mountLibrary(main,api,parentScope,{positiveId,savedState=null,on
             const hidden=hiddenSave!==null && !next.loading && !next.error && !next.hasMore && !next.items.some(item=>item.id===hiddenSave)?' The saved prompt is hidden by the current filters.':'';
             message.textContent=(next.error?'Prompt library could not be loaded. Retry.':next.loading?'Loading prompt library…':!next.items.length?'No matching prompts. Try changing the filters.':!next.hasMore?'You’re all caught up.':'')+hidden;
             message.setAttribute('role',next.error?'alert':'status');retry.hidden=!next.error;more.hidden=!next.hasMore || !!next.error;more.setAttribute('aria-disabled',String(next.loading));
-            schedule();scheduleLoad();
+            // #464: a loaded page is laid out now, not on the next frame: a starved renderer (a background tab, headless
+            // WebKit under load) can hold frames for seconds, which left a loaded shelf empty. Measurements still flush per frame.
+            if(batch && grid.isConnected) {cancelAnimationFrame(frame);flush();} else schedule();
+            scheduleLoad();
         }
         const stream=createLibraryStream({api,onChange:render,restore:restorable?{filters,state:restorable.state}:undefined});
         function maybeLoad() {
@@ -417,15 +423,26 @@ export function mountLibrary(main,api,parentScope,{positiveId,savedState=null,on
             if(Object.entries(next).every(([key,item])=>filters[key]===item)) return;
             Object.assign(filters,next);reset();
         }
-        /** "Refine this view" with the Library's one extra filter, Tags (D14); Clear also clears the toolbar category. */
+        /**
+         * "Refine this view" with the Library's filters: the concept's Folder (#444 D150, its options read when the dialog
+         * opens), Tags (D14) and Tag match (D149, D150); Clear also clears the toolbar category.
+         */
         function openFilters(trigger) {
-            openRefineDialog({scope,trigger,fields:[{key:'tags'}],values:{tags:filters.tags.join(', ')},onApply:applyRefinement,
-                onClear:cleared=>{toolbar.setValue({category:cleared.category});applyRefinement(cleared);}});
+            const refine=openRefineDialog({scope,trigger,fields:[{key:'folder',options:libraryFolderOptions([],filters.folder)},{key:'tags'},
+                {key:'tagMode',options:TAG_MATCH_OPTIONS}],values:{folder:filters.folder,tags:filters.tags.join(', '),tagMode:filters.tagMode},
+                onApply:applyRefinement,onClear:cleared=>{toolbar.setValue({category:cleared.category});applyRefinement(cleared);}});
+            folderRequest?.abort();const request=folderRequest=new AbortController();
+            // A failed read leaves "All output folders" and the current folder; nothing else depends on it.
+            api.get('/prompt_manager/library/options?include=folders',{signal:request.signal}).then(data=>{
+                if(scope.signal.aborted || folderRequest!==request || !Array.isArray(data?.folders) || data.folders.some(folder=>typeof folder!=='string')) return;
+                refine.setOptions('folder',libraryFolderOptions(data.folders,filters.folder));
+            },()=>{});
         }
-        function applyRefinement({tags,category=filters.category}) {
-            const next=[...new Set(tags.split(',').map(tag=>tag.trim()).filter(Boolean))];
-            if(category===filters.category && next.length===filters.tags.length && next.every((tag,index)=>tag===filters.tags[index])) return;
-            filters.category=category;filters.tags=next;reset();
+        function applyRefinement({tags,tagMode='all',folder='',category=filters.category}) {
+            // "Prompts without tags" takes no Tags (the dialog disables them).
+            const next=tagMode==='untagged'?[]:[...new Set(tags.split(',').map(tag=>tag.trim()).filter(Boolean))];
+            if(category===filters.category && tagMode===filters.tagMode && folder===filters.folder && next.length===filters.tags.length && next.every((tag,index)=>tag===filters.tags[index])) return;
+            filters.category=category;filters.tags=next;filters.tagMode=tagMode;filters.folder=folder;reset();
         }
         scope.listen(more,'click',()=>stream.loadMore());
         // Retry also re-reads a category list that failed (D15: Refresh is gone).
@@ -438,7 +455,7 @@ export function mountLibrary(main,api,parentScope,{positiveId,savedState=null,on
             // A write while mounted makes this stream stale: keep only what still means something.
             // A saved view is not the plain shelf, so it never replaces the Library return state (#249 M2, as Favorites).
             if(!saved) shelfVisit=mountedGeneration!==writeGeneration?{filters:savedFilters,anchor:savedAnchor}:state?{filters:savedFilters,state,scrollY:window.scrollY,layout:{inlineSize,columns:columnCount,heights:[...heights],anchor:savedAnchor}}:null;
-            toolbarScope.dispose();stream.dispose();optionsRequest?.abort();observer?.disconnect();measurementObserver?.disconnect();cancelAnimationFrame(frame);cancelAnimationFrame(loadFrame);for(const card of cards.values())removeCard(card);cards.clear();});
+            toolbarScope.dispose();stream.dispose();optionsRequest?.abort();folderRequest?.abort();observer?.disconnect();measurementObserver?.disconnect();cancelAnimationFrame(frame);cancelAnimationFrame(loadFrame);for(const card of cards.values())removeCard(card);cards.clear();});
         loadCategories();if(restorable) {render(stream.state);if(restorable.state.loading || restorable.state.error)stream.loadMore();} else reset();
     }
 

web/js/workspace/media-viewer.mjs

@@ -3,6 +3,7 @@ import {createViewerSession} from './viewer-state.mjs';
 import {rootedFileAdapter} from './inspection-state.mjs';
 import {mountFileInspector} from './file-inspector.mjs';
 import {displayPreferences, readPreferences} from './preferences.mjs';
+import {createWorkspaceIcon} from './icons.mjs';
 
 let viewerId = 0;
 // Upper bound on waiting for a closed Viewer's in-flight image view before destroying it.
@@ -39,7 +40,7 @@ export function createMediaViewer({api,container,parentScope,inspectionControlle
     function open(options) {
         if (scope.signal.aborted) return;
         active?.close(false);
-        const {items,selectedKey,trigger = document.activeElement} = options;
+        const {items,selectedKey,trigger = document.activeElement,details:detailsMode} = options;
         let mounted = false, initial, entry, currentKey, viewer = null, withoutViewer = false, mediaScope = null, closed = false;
         let originalDownload = null, inspectionState = inspectionController.state, collectionChanges = null;
         const session = createViewerSession({adapter,items,initialKey:selectedKey,onChange:(state,selected) => {
@@ -56,7 +57,10 @@ export function createMediaViewer({api,container,parentScope,inspectionControlle
         const heading = element('div'), title = element('h2');title.id = 'viewer-title';
         heading.append(title,element('p','muted small','One viewer · image, prompts, generation data, workflow'));
         const close = createButton('×');close.id = 'viewer-close';close.setAttribute('aria-label','Close image viewer');close.autofocus = true;
-        header.append(heading,close);
+        // #450 (D156): the view switch before ✕ names the view it switches to; the stored default picks the first view.
+        let fullscreen = displayPreferences()?.viewerOpenMode === 'fullscreen';
+        const mode = createButton('','icon-btn');mode.id = 'viewer-mode';
+        header.append(heading,mode,close);
         const body = element('div','viewer-layout'), left = element('div','viewer-left');
         const stage = element('div');stage.id = 'viewer-stage';stage.tabIndex = 0;
         stage.setAttribute('role','region');stage.setAttribute('aria-label','Media preview');
@@ -79,27 +83,37 @@ export function createMediaViewer({api,container,parentScope,inspectionControlle
         local.listen(download,'click',event => {session.stop();if (download.getAttribute('aria-disabled') === 'true') event.preventDefault();});
         controls.append(download);
         const next = button('Next media','Next →',() => session.navigate(1));
-        // D97: "Collapse viewer details by default" opens the viewer with its details hidden behind one toggle.
-        const collapsed = displayPreferences()?.sidebarCollapsedByDefault === true;
-        let detailsToggle = null;
-        if (collapsed) {
-            detailsToggle = createButton('Show details');detailsToggle.setAttribute('aria-expanded','false');
-            detailsToggle.setAttribute('aria-controls','viewer-details');controls.append(detailsToggle);
-        }
+        // D97: "Collapse viewer details by default" opens the viewer with its details hidden behind one toggle. D155
+        // (#448): `details: 'shown'` (Inspect a file) keeps the toggle but opens with the details showing. D156 (#450):
+        // the toggle is also there while the view is full screen.
+        const collapsible = displayPreferences()?.sidebarCollapsedByDefault === true, collapsed = collapsible && detailsMode !== 'shown';
+        const detailsToggle = createButton('');detailsToggle.setAttribute('aria-controls','viewer-details');
         const position = element('p','media-position muted');position.id = `media-viewer-position-${++viewerId}`;
         position.setAttribute('role','status');
         const reason = element('p','media-reason muted');reason.id = `media-viewer-reason-${viewerId}`;
         const slideshowReason = element('p','media-slideshow-reason muted');slideshowReason.id = `media-viewer-slideshow-${viewerId}`;
         controls.append(position,reason,slideshowReason);left.append(stage,controls);
         const details = element('aside');details.id = 'viewer-details';details.setAttribute('aria-label','Image metadata');
-        if (detailsToggle) {
-            details.hidden = true;
-            local.listen(detailsToggle,'click',() => {
-                details.hidden = !details.hidden;
-                detailsToggle.textContent = details.hidden ? 'Show details' : 'Hide details';
-                detailsToggle.setAttribute('aria-expanded',String(!details.hidden));
-            });
+        details.hidden = collapsed;
+        function paintToggle() {
+            detailsToggle.textContent = details.hidden ? 'Show details' : 'Hide details';
+            detailsToggle.setAttribute('aria-expanded',String(!details.hidden));
         }
+        /** The toggle is in the toolbar only in full screen or with the collapse setting; without both the details show. */
+        function syncToggle() {
+            if (collapsible || fullscreen) { if (!detailsToggle.isConnected) controls.insertBefore(detailsToggle,position); }
+            else { detailsToggle.remove();details.hidden = false; }
+            paintToggle();
+        }
+        function paintMode() {
+            if (fullscreen) dialog.classList.add('viewer-fullscreen'); else dialog.classList.remove('viewer-fullscreen');
+            const label = fullscreen ? 'Centered view' : 'Full screen view';
+            mode.setAttribute('aria-label',label);mode.title = label;
+            mode.replaceChildren(createWorkspaceIcon(fullscreen ? 'minimize' : 'maximize'));
+        }
+        local.listen(detailsToggle,'click',() => {details.hidden = !details.hidden;paintToggle();});
+        local.listen(mode,'click',() => {fullscreen = !fullscreen;paintMode();syncToggle();});
+        paintMode();syncToggle();
         body.append(left,details);dialog.append(header,body);container.append(dialog);
         const localSender = {open(roles,trigger) {session.stop();sender.open(roles,trigger);}};
         const inspector = mountFileInspector(details,{api,sender:localSender,parentScope:local,variant:'full',onInteraction:() => session.stop(),onRetry:() => inspectionController.retry(),

web/js/workspace/preferences.mjs

@@ -21,7 +21,7 @@
  */
 export const PREFERENCES_KEY = 'gallerySettings';
 export const WORKSPACE_PREFERENCE_KEYS = Object.freeze(['workspaceLayout', 'workspaceDensity', 'autoLoadMetadata', 'cacheMetadata',
-    'imageQuality', 'videoAutoplay', 'videoMute', 'videoLoop', 'showFilePaths', 'showImageInfo', 'sidebarCollapsedByDefault']);
+    'imageQuality', 'videoAutoplay', 'videoMute', 'videoLoop', 'showFilePaths', 'showImageInfo', 'sidebarCollapsedByDefault', 'viewerOpenMode']);
 
 // The changes this tab could not store, overlaid on every read until storage accepts them.
 const tabOnly = new Map();

web/js/workspace/saved-views.mjs

@@ -47,6 +47,9 @@ const FIELDS = Object.freeze({
     }),
     library: Object.freeze({
         q: text(500), category: text(100), sort: choice(['newest', 'oldest', 'name'], 'newest'), tags,
+        // #444: how the Tags match ('untagged' never goes with tags; the listing refuses both) and the output folder
+        // ('/'-joined, below a root).
+        tag_mode: choice(['all', 'any', 'untagged'], 'all'), folder: text(1024),
     }),
 });
 
@@ -63,6 +66,8 @@ export function normalizeViewState(page, state) {
     }));
     // Favorites and a collection are never both a view's filter: the collection wins, as the listing does (D140).
     if (normalized.favorites === true && normalized.collection) normalized.favorites = false;
+    // Untagged and tags are never both a view's filter either: the tags win, as the server's normalizer does.
+    if (normalized.tag_mode === 'untagged' && normalized.tags.length) normalized.tag_mode = 'all';
     return normalized;
 }
 
@@ -113,10 +118,12 @@ export function imagesViewState(filters) {
 /** Library shelf filters (library.mjs `filters`) from a saved state, and the state of the current filters. */
 export function libraryFilters(state) {
     const view = normalizeViewState('library', state);
-    return { text: view.q, category: view.category, sort: view.sort, tags: view.tags };
+    return { text: view.q, category: view.category, sort: view.sort, tags: view.tags, tagMode: view.tag_mode, folder: view.folder };
 }
 export function libraryViewState(filters) {
-    return normalizeViewState('library', { q: filters.text, category: filters.category, sort: filters.sort, tags: filters.tags });
+    return normalizeViewState('library', { q: filters.text, category: filters.category, sort: filters.sort, tags: filters.tags,
+        // A filter the shelf lacks (undefined) takes its key's default in normalizeViewState.
+        tag_mode: filters.tagMode, folder: filters.folder });
 }
 
 /**

web/js/workspace/system-appearance.mjs

@@ -13,7 +13,10 @@
  *   (`showImageInfo`), "Collapse viewer details by default" (`sidebarCollapsedByDefault`) and the operation "Reset
  *   display defaults", which drafts the gallery's defaults for these browser rows and touches no other key.
  *
- * - #317: a second surface, "Thumbnails", holds the server's thumbnail format, quality and size limit
+ * - #450 (D156, amends D97): a "Viewer" surface after it holds "Open viewer in" (Centered→`centered`, Full
+ *   screen→`fullscreen`, `viewerOpenMode`) and the moved "Collapse viewer details by default"; "Reset display defaults"
+ *   covers both surfaces' browser rows.
+ * - #317: a further surface, "Thumbnails", holds the server's thumbnail format, quality and size limit
  *   (system-thumbnails.mjs).
  *
  * Controls follow the store: a value read later (the host settings, another tab's change) repaints them. A host row
@@ -50,14 +53,20 @@ export const APPEARANCE_ROWS = Object.freeze([
         { type: 'select', options: [['newtab', 'New tab'], ['popup', 'Popup window']] }],
     ['timeout', 'Node search auto-hide (seconds)', 'Node search behavior only. Use 0 to keep results visible.', 'host', 'result_timeout',
         { type: 'number' }],
-    // D97: the legacy gallery's card and viewer settings.
+    // D97: the legacy gallery's card setting (its viewer setting moved to VIEWER_ROWS, D156).
     ['card-details', 'Show image details on cards', 'Show each card’s title and file details under the preview.', 'browser', 'showImageInfo', BOOLEAN],
+]);
+
+/** #450 (D156): the "Viewer" surface's rows, in the same shape. */
+export const VIEWER_ROWS = Object.freeze([
+    ['viewer-mode', 'Open viewer in', 'Full screen fills the browser window; Centered is the framed viewer.', 'browser', 'viewerOpenMode',
+        { type: 'select', options: [['centered', 'Centered'], ['fullscreen', 'Full screen']] }],
     ['viewer-details', 'Collapse viewer details by default', 'Open the image viewer with its details panel hidden.', 'browser',
         'sidebarCollapsedByDefault', BOOLEAN],
 ]);
 
-/** The browser keys "Reset display defaults" restores (D97). */
-export const DISPLAY_KEYS = Object.freeze(APPEARANCE_ROWS.filter(row => Array.isArray(row) && row[3] === 'browser').map(row => row[4]));
+/** The browser keys "Reset display defaults" restores (D97), on both surfaces. */
+export const DISPLAY_KEYS = Object.freeze([...APPEARANCE_ROWS, ...VIEWER_ROWS].filter(row => Array.isArray(row) && row[3] === 'browser').map(row => row[4]));
 
 /** A whole number of seconds from 0 to 300 (the server's `validate_result_timeout`), else null. */
 export function resultTimeout(text) {
@@ -104,29 +113,36 @@ function paint(input, control, value) {
 
 /**
  * Mount Appearance in `container` (System's `.settings-layout`); returns the section's nodes: "Browsing & display",
- * then the server's thumbnail settings (#317, system-thumbnails.mjs).
+ * "Viewer" (D156), then the server's thumbnail settings (#317, system-thumbnails.mjs).
  */
 export function mountAppearance(container, { scope, drafts, api, activity }) {
     const surface = el('section', undefined, 'surface');
     surface.append(el('h2', 'Browsing & display'), el('p', 'One set of defaults; toolbar changes apply to the current view.'));
     const controls = [];
-    for (const row of APPEARANCE_ROWS) {
-        if (row === 'continuous') { surface.append(continuousRow()); continue; }
-        const [id, label, description, domain, key, control] = row;
-        const onChange = value => {
-            if (key === 'result_timeout') {
-                const seconds = resultTimeout(input.value);
-                if (seconds === null) { drafts.reject(domain, key); return; }
-                value = seconds;
-            }
-            drafts.set(domain, key, value);
-        };
-        const node = settingRow({ id, label, description, control, scope, onChange });
-        surface.append(node);
-        const input = controlOf(node, id);
-        input.dataset.key = key;
-        controls.push({ input, control, domain, key, text: node.children[0], note: null });
+    function addRows(target, rows) {
+        for (const row of rows) {
+            if (row === 'continuous') { target.append(continuousRow()); continue; }
+            const [id, label, description, domain, key, control] = row;
+            const onChange = value => {
+                if (key === 'result_timeout') {
+                    const seconds = resultTimeout(input.value);
+                    if (seconds === null) { drafts.reject(domain, key); return; }
+                    value = seconds;
+                }
+                drafts.set(domain, key, value);
+            };
+            const node = settingRow({ id, label, description, control, scope, onChange });
+            target.append(node);
+            const input = controlOf(node, id);
+            input.dataset.key = key;
+            controls.push({ input, control, domain, key, text: node.children[0], note: null });
+        }
     }
+    addRows(surface, APPEARANCE_ROWS);
+    // D156: the viewer's defaults in their own surface.
+    const viewer = el('section', undefined, 'surface');
+    viewer.append(el('h2', 'Viewer'), el('p', 'How the image viewer opens; the button beside ✕ switches the current view.'));
+    addRows(viewer, VIEWER_ROWS);
     surface.append(operationRow({ title: 'Reset display defaults',
         description: 'Restore this browser’s display defaults, then save. Other settings are kept.', label: 'Reset', scope,
         onClick: () => {
@@ -146,5 +162,5 @@ export function mountAppearance(container, { scope, drafts, api, activity }) {
     };
     scope.cleanup(drafts.subscribe(repaint));
     repaint();
-    return [surface, mountThumbnailSettings(container, { scope, drafts, api, activity })];
+    return [surface, viewer, mountThumbnailSettings(container, { scope, drafts, api, activity })];
 }

web/js/workspace/system-sources.mjs

@@ -14,6 +14,9 @@
  *   it to the host draft of `gallery_root_paths`; Remove takes one out. Both wait for the one Save row.
  * - "Watch for new outputs" (`monitoring_enabled`) and "Worker threads" (`worker_threads`, a range up to the server's
  *   `cpu_count`, D113) are host drafts too. "Enrich existing metadata" reviews a scan of every source.
+ * - D158 (#247): "Save prompts from other workflows" (`save_foreign_runs`), a host draft below "Watch for new outputs",
+ *   applied by the server at once (the monitor reads it for every file); disabled, saying why, while monitoring is off,
+ *   unless the running monitor still watches until a restart (D128 `stopping`): then it can still be set.
  * - "Thumbnail generation → Review options" opens "Thumbnail maintenance" (thumbnail-review.mjs, S5 Task 14), which
  *   starts the thumbnail Activity job.
  *
@@ -63,6 +66,21 @@ export const LOAD_FAILED_TEXT = 'Sources could not be loaded. Your saved data ha
 export const DUPLICATE_TEXT = 'This folder is already a source.';
 export const VALIDATE_FAILED_TEXT = 'The folder could not be checked. Try again.';
 export const REFRESH_FAILED_TEXT = 'Files could not be refreshed. Try again.';
+/** D158 (#247): the row the concept does not draw, below "Watch for new outputs". */
+export const FOREIGN_RUNS = Object.freeze({ id: 'foreign-runs', key: 'save_foreign_runs', label: 'Save prompts from other workflows',
+    description: 'New outputs of workflows without a PromptManager node add their prompt to the library when it is certain.',
+    off: 'Applies while Watch for new outputs is on.' });
+
+/**
+ * D158: the row for the drafted `save_foreign_runs` (`value`) and `monitoring_enabled` (`monitoring`): disabled until
+ * read, and while monitoring is off, when `note` says why (the value is kept and saved as it is). `stopping`: a source
+ * is still watched until the restart (D128), so the value can still be set; the note stays.
+ */
+export function foreignRunsState({ value, monitoring, stopping = false }) {
+    if (value === undefined) return { disabled: true, checked: null, note: null };
+    const off = monitoring === false;
+    return { disabled: off && !stopping, checked: value === true, note: off ? FOREIGN_RUNS.off : null };
+}
 
 // Labels of the sources added in this page session (validate-source answers them), for the rows not saved yet.
 const addedLabels = new Map();
@@ -267,13 +285,16 @@ export function mountSources(container, { api, activity, scope, drafts, notify =
     const monitoring = settingRow({ id: 'monitoring', label: 'Watch for new outputs',
         description: 'New files enter the same library. Existing scans do not count as new runs.', control: { type: 'toggle' },
         scope, onChange: value => drafts.set('host', 'monitoring_enabled', value) });
+    const foreign = settingRow({ id: FOREIGN_RUNS.id, label: FOREIGN_RUNS.label, description: FOREIGN_RUNS.description,
+        control: { type: 'toggle' }, scope, onChange: value => drafts.set('host', FOREIGN_RUNS.key, value) });
     const workers = settingRow({ id: 'workers', label: 'Worker threads',
         description: 'Balance background scans and thumbnail work against generation.', control: { type: 'range', value: 4 },
         scope, onChange: value => drafts.set('host', 'worker_threads', value) });
-    const hostRows = [[monitoring, 'monitoring', 'monitoring_enabled'], [workers, 'workers', 'worker_threads']]
+    const hostRows = [[monitoring, 'monitoring', 'monitoring_enabled'], [foreign, FOREIGN_RUNS.id, FOREIGN_RUNS.key],
+        [workers, 'workers', 'worker_threads']]
         .map(([row, id, key]) => ({ input: controlOf(row, id), key, text: row.children[0], note: null }));
     for (const row of hostRows) row.input.dataset.key = row.key;
-    surface.append(monitoring, workers,
+    surface.append(monitoring, foreign, workers,
         operationRow({ scope, title: 'Enrich existing metadata', description: 'Resolve both prompts, deduplicate recipes, preserve uncertain links.',
             label: 'Review scan', onClick: scanAll }),
         operationRow({ scope, title: 'Thumbnail generation', description: 'Generate missing previews; rebuilding all is a separate explicit option.',
@@ -348,6 +369,10 @@ export function mountSources(container, { api, activity, scope, drafts, notify =
         for (const row of hostRows) {
             const value = drafts.get('host', row.key);
             row.input.disabled = value === undefined;
+            if (row.key === FOREIGN_RUNS.key) {
+                paintForeign(row, value, unavailable);
+                continue;
+            }
             if (row.key === 'worker_threads') {
                 // D113: the range ends at the server's core count (the concept's fixed 16).
                 const cpus = drafts.get('host', 'cpu_count');
@@ -360,6 +385,24 @@ export function mountSources(container, { api, activity, scope, drafts, notify =
         }
     }
 
+    // D158: disabled while monitoring is drafted off, with the reason under the description.
+    function paintForeign(row, value, unavailable) {
+        const stopping = sources?.effective?.some(source => source.watch === 'stopping') === true;
+        const state = foreignRunsState({ value, monitoring: drafts.get('host', 'monitoring_enabled'), stopping });
+        row.input.disabled = state.disabled;
+        if (state.checked !== null) row.input.checked = state.checked;
+        const note = unavailable && value === undefined ? HOST_UNAVAILABLE : state.note;
+        if (note && row.note?.textContent !== note) {
+            row.note?.remove();
+            row.note = el('p', note, 'small');
+            row.note.id = `${FOREIGN_RUNS.id}-note`;
+            row.text.append(row.note);
+        } else if (!note && row.note) { row.note.remove(); row.note = null; }
+        // The note is the switch's description, so a screen reader says why it is disabled.
+        if (row.note) row.input.setAttribute('aria-describedby', row.note.id);
+        else row.input.removeAttribute('aria-describedby');
+    }
+
     // The sources changed since this page read them (a refused per-source scan): read them and the saved values again.
     function reread() {
         load();
@@ -378,6 +421,7 @@ export function mountSources(container, { api, activity, scope, drafts, notify =
         if (answer === STALE || scope.signal.aborted) return;
         sources = answer?.sources ?? null;
         paintList();
+        paintControls();
     }
 
     // A save (or a reread after a conflict) changed the saved sources or monitoring: read their state again. The first

web/js/workspace/vocabulary-state.mjs

@@ -26,6 +26,12 @@ const UNCONFIRMED_STATUS = new Set([502, 503, 504]);
 // Phases that may send a command; done, storage-changed and unknown-outcome need reset().
 const READY = new Set(['idle', 'previewed', 'write-error', 'conflict']);
 const IMPACT_FIELDS = ['positive_prompts', 'other_prompts', 'negative_variants', 'archived_negative_variants'];
+/** #444 tag sorts as [value, label]: `created` (today's order) first, then the legacy tag manager's four. */
+export const TAG_SORTS = Object.freeze([['created', 'Oldest first'], ['name', 'Name A–Z'], ['name-desc', 'Name Z–A'],
+    ['prompts', 'Most prompts'], ['prompts-asc', 'Fewest prompts']].map(pair => Object.freeze(pair)));
+// The route's cap on a tag search (database/vocabulary.py _MAX_SEARCH).
+export const TAG_SEARCH_MAX = 200;
+const DEFAULT_QUERY = Object.freeze({search: '', sort: 'created'});
 
 const object = value => value !== null && typeof value === 'object' && !Array.isArray(value);
 const id = value => Number.isSafeInteger(value) && value > 0;
@@ -45,9 +51,23 @@ function storageChanged() {
 // Duck-typed by name: this ES module cannot import the UMD api-client.js.
 function isApiError(error) {return error instanceof Error && error.name === 'ApiError' && Number.isInteger(error.status);}
 
-function validatePage(page, kind, cursor) {
+/** A tags query, trimmed; categories take none. Throws on anything the route would refuse. */
+function tagQuery(kind, query) {
+    if (query === undefined) return DEFAULT_QUERY;
+    requireValue(kind === 'tags' && object(query), 'Only tags take a search and sort');
+    const {search = '', sort = 'created'} = query;
+    requireValue(typeof search === 'string' && search.trim().length <= TAG_SEARCH_MAX, 'Invalid tag search');
+    requireValue(TAG_SORTS.some(([value]) => value === sort), 'Invalid tag sort');
+    return Object.freeze({search: search.trim(), sort});
+}
+
+function validatePage(page, kind, cursor, query) {
     const rows = page?.[kind];
     const {key, valid} = KINDS[kind];
+    // A search or sort the server did not echo was not applied (an older server): never show that list as filtered.
+    const plain = query.search === '' && query.sort === 'created' && page?.search === undefined && page?.sort === undefined;
+    const applied = plain || (page?.search === query.search && page?.sort === query.sort);
+    requireValue(applied && optionalCount(page?.untagged_count), `Unconfirmed ${kind} search`);
     requireValue(object(page) && page.success === true && epochOf(page.library_epoch) && Array.isArray(rows) &&
         rows.every(row => object(row) && valid(row) && count(row.prompt_count)) &&
         new Set(rows.map(key)).size === rows.length && count(page.total) && typeof page.has_more === 'boolean' &&
@@ -57,12 +77,18 @@ function validatePage(page, kind, cursor) {
 }
 
 /** One bounded, append-only list of tags or categories. */
-export function createVocabularyStream({api, kind, onChange = () => {}}) {
+/**
+ * One bounded, append-only list of tags or categories. A tags stream takes `query` (`{search, sort}`, #444) and
+ * `reset(query)` replaces it; `state.untaggedCount` is the last count a tags page reported (kept across resets so the
+ * Untagged prompts row does not flicker), `null` before any.
+ */
+export function createVocabularyStream({api, kind, onChange = () => {}, query: initialQuery}) {
     requireValue(Object.hasOwn(KINDS, kind), 'Unknown vocabulary kind');
     const {key} = KINDS[kind];
-    const initial = () => ({items: Object.freeze([]), total: 0, cursor: null, hasMore: true, loading: false, error: null,
-        epoch: null, resetRequired: false});
-    let state = Object.freeze(initial());
+    let query = tagQuery(kind, initialQuery);
+    const initial = untaggedCount => ({items: Object.freeze([]), total: 0, cursor: null, hasMore: true, loading: false, error: null,
+        epoch: null, resetRequired: false, query, untaggedCount});
+    let state = Object.freeze(initial(null));
     let request = null, disposed = false, seen = new Set();
     const emit = patch => {state = Object.freeze({...state, ...patch}); onChange(state);};
 
@@ -70,13 +96,16 @@ export function createVocabularyStream({api, kind, onChange = () => {}}) {
         if (disposed || state.loading || !state.hasMore || state.resetRequired) return 0;
         const current = request = new AbortController();
         const cursor = state.cursor;
-        const query = new URLSearchParams({limit: String(LIMIT)});
-        if (cursor !== null) query.set('cursor', cursor);
+        const asked = query;
+        const params = new URLSearchParams({limit: String(LIMIT)});
+        if (asked.search) params.set('search', asked.search);
+        if (asked.sort !== 'created') params.set('sort', asked.sort);
+        if (cursor !== null) params.set('cursor', cursor);
         emit({loading: true, error: null});
         try {
-            const page = await api.get(`${BASE}/${kind}?${query}`, {signal: current.signal, cache: 'no-store'});
+            const page = await api.get(`${BASE}/${kind}?${params}`, {signal: current.signal, cache: 'no-store'});
             if (disposed || current !== request) return 0;
-            validatePage(page, kind, cursor);
+            validatePage(page, kind, cursor, asked);
             request = null;
             if (state.epoch !== null && page.library_epoch !== state.epoch) {
                 emit({loading: false, error: storageChanged(), resetRequired: true});
@@ -85,7 +114,8 @@ export function createVocabularyStream({api, kind, onChange = () => {}}) {
             const incoming = page[kind].filter(row => !seen.has(key(row))).map(row => Object.freeze({...row}));
             incoming.forEach(row => seen.add(key(row)));
             emit({items: incoming.length ? Object.freeze([...state.items, ...incoming]) : state.items, total: page.total,
-                cursor: page.next_cursor, hasMore: page.has_more, loading: false, error: null, epoch: page.library_epoch});
+                cursor: page.next_cursor, hasMore: page.has_more, loading: false, error: null, epoch: page.library_epoch,
+                untaggedCount: page.untagged_count ?? state.untaggedCount});
             return incoming.length;
         } catch (error) {
             if (!disposed && current === request) {request = null; emit({loading: false, error});}
@@ -96,13 +126,14 @@ export function createVocabularyStream({api, kind, onChange = () => {}}) {
     return {
         get state() {return state;},
         loadMore,
-        /** Drop every loaded row and read again from the first page. */
-        reset() {
+        /** Drop every loaded row and read again from the first page, with `next` as the new query when given. */
+        reset(next) {
+            if (next !== undefined) query = tagQuery(kind, next);
             if (disposed) return Promise.resolve(0);
             request?.abort();
             request = null;
             seen = new Set();
-            emit(initial());
+            emit(initial(state.untaggedCount));
             return loadMore();
         },
         dispose() {disposed = true; request?.abort(); request = null;},
@@ -254,6 +285,11 @@ export function browseFilters(kind, value) {
         : {text: '', category: value, tags: [], sort: 'newest'};
 }
 
+/** Library filters for the Untagged prompts row's Browse (#444): parents that no tag's Browse returns. */
+export function untaggedBrowseFilters() {
+    return {text: '', category: '', tags: [], sort: 'newest', tagMode: 'untagged'};
+}
+
 /**
  * A category row's count (#332 D143): "N positive prompts" when no image holds it as its own category, "N positive
  * prompts · M images" when both do, "M images" when only images do (stored rows, files on disk or not).

web/js/workspace/vocabulary.mjs

@@ -4,12 +4,18 @@
  * Composition, labels and order follow the approved concept: Auto-tag images
  * opens the Tag image dialog for the image selected in Images (aria-disabled
  * without one, D115) and the Images cells count the files carrying each tag
- * (#249 C2). The recorded deviation left is bounded lists with Load more. Rows
- * are real vocabulary pages from vocabulary-state.mjs; Browse hands the exact
- * constraint to the existing Library filters.
+ * (#249 C2). The recorded deviations are bounded lists with Load more and (D149,
+ * #444) the Tag vocabulary's Untagged prompts row and its tag search and sort,
+ * drawn with the concept's `.operation` and toolbar tokens. Rows are real
+ * vocabulary pages from vocabulary-state.mjs; Browse hands the exact constraint
+ * to the existing Library filters.
  */
 import { createButton } from '../shared/toolkit/core.js';
-import { browseFilters, categoryBrowseTarget, categoryRowText, createVocabularyCommand, createVocabularyStream } from './vocabulary-state.mjs';
+import { TAG_SEARCH_MAX, TAG_SORTS, browseFilters, categoryBrowseTarget, categoryRowText, createVocabularyCommand, createVocabularyStream,
+    untaggedBrowseFilters } from './vocabulary-state.mjs';
+import { SEARCH_DELAY } from './toolbar.mjs';
+import { createWorkspaceIcon } from './icons.mjs';
+import { countOf } from './plural.mjs';
 import { browseLibrary, invalidateLibraryAfterWrite } from './library.mjs';
 import { createVocabularyDialog } from './vocabulary-dialog.mjs';
 import { pageHeading } from './heading.mjs';
@@ -20,6 +26,9 @@ import { showWorkspaceToast } from './toast.mjs';
 /** D115: why Auto-tag images (and System › Auto-tagging's Preview tagging review) only explains without a selection. */
 export const NO_SELECTION = 'Select an image in Images first.';
 
+const NO_MATCH = 'No tags match this search.';
+// The Untagged prompts row before the first tags page reports its count.
+const UNTAGGED_PENDING = 'Positive prompts without a tag';
 const COPY = {
     tags: {empty: 'No tags yet.', error: 'Tags could not be loaded. Nothing was changed.', more: 'Load more tags'},
     categories: {empty: 'No categories yet.', error: 'Categories could not be loaded. Nothing was changed.', more: 'Load more categories'},
@@ -75,7 +84,7 @@ function categoryRow(category) {
  * in view (stacked layout below 700px), pages do not chain on their own. Before any row arrives (loading, failed) the list shows no rows, so a
  * failed read never looks like zero counts.
  */
-function mountList(scope, api, kind, {content, rowsHost, renderRow, after}) {
+function mountList(scope, api, kind, {content, rowsHost, renderRow, after, query}) {
     const copy = COPY[kind];
     const empty = el('p', copy.empty, 'empty');
     const continuation = el('div', undefined, 'vocabulary-continuation');
@@ -115,6 +124,7 @@ function mountList(scope, api, kind, {content, rowsHost, renderRow, after}) {
         const failed = Boolean(state.error);
         content.hidden = !rendered;
         empty.hidden = Boolean(rendered || state.loading || failed || state.hasMore);
+        empty.textContent = state.query?.search ? NO_MATCH : copy.empty;
         status.textContent = state.loading ? 'Loading…' : '';
         status.hidden = !state.loading;
         alert.hidden = !failed;
@@ -129,7 +139,7 @@ function mountList(scope, api, kind, {content, rowsHost, renderRow, after}) {
         continuation.hidden = !(state.loading || failed || state.hasMore);
         for (const listener of listeners) listener(state);
     }
-    const stream = createVocabularyStream({api, kind, onChange: render});
+    const stream = createVocabularyStream({api, kind, onChange: render, query});
     scope.listen(more, 'click', () => stream.loadMore());
     scope.listen(retry, 'click', () => (stream.state.resetRequired ? stream.reset() : stream.loadMore()));
     let observer = null;
@@ -156,6 +166,38 @@ function mountList(scope, api, kind, {content, rowsHost, renderRow, after}) {
     return {stream, itemOf: row => items.get(row), rowFor, subscribe};
 }
 
+/**
+ * The tag search and sort (D149): the concept toolbar's `search.toolbar` > `label.search-field` (search icon and a
+ * search input) and a `select`, built here because the shared toolbar owns the page-level ids. A visually hidden
+ * status announces the result of a changed search or sort.
+ */
+function tagTools() {
+    const root = document.createElement('search');
+    root.className = 'toolbar vocabulary-tools';
+    const label = el('label', undefined, 'search-field');
+    label.htmlFor = 'tag-search';
+    const search = document.createElement('input');
+    search.id = 'tag-search';
+    search.type = 'search';
+    search.placeholder = 'Search tags…';
+    search.maxLength = TAG_SEARCH_MAX;
+    search.setAttribute('aria-label', 'Search tags');
+    label.append(createWorkspaceIcon('search'), search);
+    const sort = document.createElement('select');
+    sort.id = 'tag-sort';
+    sort.setAttribute('aria-label', 'Sort tags');
+    for (const [value, text] of TAG_SORTS) {
+        const option = el('option', text);
+        option.value = value;
+        sort.append(option);
+    }
+    const status = el('span', '', 'ca-visually-hidden');
+    status.setAttribute('role', 'status');
+    root.append(label, sort, status);
+    return {root, search, sort, read: () => ({search: search.value, sort: sort.value}),
+        announce: text => { if (status.textContent !== text) status.textContent = text; }};
+}
+
 /** Mount the screen into `main` below the shell's heading; `scope` owns every request, listener and observer. */
 export function mountVocabulary(main, api, scope) {
     main.classList.add('vocabulary-main');
@@ -188,8 +230,11 @@ export function mountVocabulary(main, api, scope) {
     const body = el('tbody');
     table.append(el('caption', 'Reusable tags'), head, body);
     wrap.append(table);
+    // D149 (#444): the parents no tag's Browse returns, as a concept `.operation` row, then the tag search and sort.
+    const untagged = operation('Untagged prompts', UNTAGGED_PENDING, 'Browse');
+    const tools = tagTools();
     tagSurface.append(el('h2', 'Tag vocabulary'),
-        el('p', 'Category/tag autocomplete uses this same vocabulary across nodes and dialogs.'), wrap);
+        el('p', 'Category/tag autocomplete uses this same vocabulary across nodes and dialogs.'), untagged.row, tools.root, wrap);
 
     const categorySurface = el('section', undefined, 'surface');
     const categoryRows = el('div', undefined, 'vocabulary-categories');
@@ -200,7 +245,17 @@ export function mountVocabulary(main, api, scope) {
     grid.append(tagSurface, categorySurface);
     main.append(grid);
 
-    const tags = mountList(scope, api, 'tags', {content: wrap, rowsHost: body, renderRow: tagRow, after: wrap});
+    const tags = mountList(scope, api, 'tags', {content: wrap, rowsHost: body, renderRow: tagRow, after: wrap, query: tools.read()});
+    let announcing = false;
+    tags.subscribe(state => {
+        if (state.untaggedCount !== null) untagged.row.querySelector('p').textContent = countOf(state.untaggedCount, 'positive prompt');
+        // A search or sort the person changed is announced once its first page lands (or fails, which the alert says).
+        // A reset first emits its empty initial state (no epoch yet), then the read: wait for the page or the failure.
+        if (!announcing || state.loading || (state.epoch === null && !state.error)) return;
+        announcing = false;
+        if (!state.error) tools.announce(state.total ? countOf(state.total, 'tag') : state.query.search ? NO_MATCH : COPY.tags.empty);
+    });
+    scope.listen(untagged.button, 'click', () => browseLibrary(untaggedBrowseFilters()));
     const categories = mountList(scope, api, 'categories', {content: categoryRows, rowsHost: categoryRows, renderRow: categoryRow, after: categoryRows});
     scope.listen(body, 'click', event => {
         const button = event.target.closest('button[data-action="browse"]');
@@ -217,11 +272,20 @@ export function mountVocabulary(main, api, scope) {
     });
 
     let settled = Promise.resolve();
-    /** Re-read both lists from page one: after any command, rows can move between pages. */
+    /** Re-read both lists from page one: after any command, rows can move between pages. The tag search and sort stay. */
     function reload() {
         settled = Promise.all([tags.stream.reset(), categories.stream.reset()]).then(() => undefined);
         return settled;
     }
+    /** A changed tag search or sort reads the tags from page one; the categories are not affected. */
+    function searchTags() {
+        const next = tools.read(), current = tags.stream.state.query;
+        if (next.search.trim() === current.search && next.sort === current.sort) return;
+        announcing = true;
+        settled = tags.stream.reset(next).then(() => undefined);
+    }
+    scope.listen(tools.search, 'input', scope.debounce(searchTags, SEARCH_DELAY));
+    scope.listen(tools.sort, 'change', searchTags);
     /** A guarded command for one dialog; any outcome that may have changed storage re-reads the lists and the Library shelf. */
     function createCommand(onChange) {
         return createVocabularyCommand({api, onChange, onInvalidate: event => {

web/js/workspace/workspace-dialogs.mjs

@@ -24,7 +24,8 @@
 import { createButton, createSwitch } from '../shared/toolkit/core.js';
 import { noun } from './plural.mjs';
 
-const FIELD_LABELS = Object.freeze({ source: 'Source', folder: 'Folder', media: 'Media', tags: 'Tags' });
+// `tagMode` (#444, D149, D150): the Library's Tag match select after Tags; its `untagged` value disables Tags.
+const FIELD_LABELS = Object.freeze({ source: 'Source', folder: 'Folder', media: 'Media', tags: 'Tags', tagMode: 'Tag match' });
 const FIELD_ORDER = Object.keys(FIELD_LABELS);
 
 let shell = null;
@@ -179,7 +180,7 @@ export function openRefineDialog({ scope, fields, values = {}, onApply, onClear,
     requireScope(scope);
     requireCallbacks({ onApply, onClear });
     if (!Array.isArray(fields) || fields.some(item => !FIELD_ORDER.includes(item?.key))) {
-        throw new TypeError('Refine fields are source, folder, media and tags');
+        throw new TypeError('Refine fields are source, folder, media, tags and tagMode');
     }
     const controls = FIELD_ORDER.flatMap(key => {
         const item = fields.find(candidate => candidate.key === key);
@@ -199,6 +200,12 @@ export function openRefineDialog({ scope, fields, values = {}, onApply, onClear,
     for (const { key, control } of controls) {
         if (key !== 'tags') control.addEventListener('change', () => wanted.set(key, control.value));
     }
+    // "Prompts without tags" takes no Tags: the field is disabled (and Apply reports it empty) while it is chosen.
+    const tagsField = controls.find(item => item.key === 'tags');
+    const tagMode = controls.find(item => item.key === 'tagMode');
+    const followTagMode = () => { if (tagsField && tagMode) tagsField.control.disabled = tagMode.control.value === 'untagged'; };
+    tagMode?.control.addEventListener('change', followTagMode);
+    followTagMode();
     const source = controls.find(item => item.key === 'source');
     const folder = controls.find(item => item.key === 'folder' && typeof item.optionsFor === 'function');
     const followSource = (chosen = folder.control.value) => {
@@ -211,7 +218,7 @@ export function openRefineDialog({ scope, fields, values = {}, onApply, onClear,
         source.control.addEventListener('change', () => { followSource(); wanted.set('folder', folder.control.value); });
     }
     const read = () => Object.fromEntries(controls.map(({ key, control }) =>
-        [key, key === 'tags' ? control.value.trim() : control.value]));
+        [key, key === 'tags' ? (control.disabled ? '' : control.value.trim()) : control.value]));
 
     const clear = createButton('Clear filters');
     const apply = createButton('Apply filters', 'ca-primary');
@@ -219,6 +226,7 @@ export function openRefineDialog({ scope, fields, values = {}, onApply, onClear,
         // The cleared Source's folders first, so the cleared Folder value is one of them.
         if (source && folder) { source.control.value = source.cleared; followSource(); }
         for (const { control, cleared } of controls) control.value = cleared;
+        followTagMode();
         const reset = read();
         finish();
         onClear({ ...reset, category: '' });

web/js/workspace/workspace.css

@@ -187,6 +187,13 @@ main:focus-visible { outline: 2px solid var(--ca-focus); outline-offset: 4px; }
 /* #249 D34: under 200% page zoom the concept's 94dvh outgrows the viewport; the dialog keeps within it. */
 .ca-tools #viewer-dialog { max-height: min(94dvh, 94%); /* D34 */ }
 #viewer-dialog .dialog-header > div { min-width: 0; }
+/* #450 D156: the view switch sits before ✕ (8px apart; the header's gap is the concept's 18px), and Full screen fills the
+   browser window edge to edge (no outer margin), beating the concept's box, D34's max-height and the ≤700px height.
+   At ≤700px the viewer is already full-bleed, so the switch hides there. */
+#viewer-dialog .dialog-header > #viewer-mode { margin-inline-start: auto; margin-inline-end: -10px; }
+.ca-tools #viewer-dialog.viewer-fullscreen[open] { inset: 0; width: auto; height: auto; max-width: none; max-height: none; margin: 0; border: 0; border-radius: 0; }
+/* `!important` only to beat the concept's `.icon-btn { display: inline-grid !important }`. */
+@media (max-width: 700px) { #viewer-dialog .dialog-header > #viewer-mode { display: none !important; } }
 #viewer-title { overflow-wrap: anywhere; /* D34 */ }
 #viewer-dialog, #viewer-dialog::backdrop { overscroll-behavior: contain; }
 #viewer-details { overscroll-behavior: contain; /* D34 */ }