DemonGatanjieuandClaude Opus 5.5
139380e960
- An answer past Antigravity's size is filed away unread (eight LoRA searches in the exam): search_models answers small (loaders once, four trigger words, a LoRA's author weight); describe_canvas gives a big canvas's last nodes a line each; mcp_fit.py keeps any other list answer within the size and names what it left out. - An empty canvas: run_workflow, describe_canvas and get_situation say to open a workflow first; get_image_info says a picture's workflow opens with open_workflow. - ask_user does not put a "test" question to the user; a task's picture count counts what its runs made. - AMB failed to load now and then: ComfyUI loads every file at once and ui_ai_workbench.js set ui_ai_toolbox.js's sideHost before it was made when the toolbox file ran first. Connected on first draw now; tests/load_order.mjs runs every file first. - input_facts moved to api/picture_mask.py to keep mcp_tools.py short. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
52 lines
2.7 KiB
Python
52 lines
2.7 KiB
Python
"""Answers an AI app reads in the chat. Past about 4,000 characters Antigravity (and apps like it) keeps
|
|
a tool's answer as a file for the AI to open, and an AI told to use only Anomalous's tools never
|
|
opens it: it sees nothing and asks again with smaller limits (the 1.63 exam: search_models eight
|
|
times for one LoRA). An answer that is one list and longer than BUDGET keeps the items that fit, in
|
|
their order, and says how many it left out, by name when its items have one, and how to reach
|
|
them. Tools that fit their answers themselves (get_situation, search_models, list_workflows) are
|
|
under it already; this is the net under the rest (list_combos, recent_activity and any list to
|
|
come). Nothing is shortened inside an item: a prompt cut in half could be put in as it is.
|
|
mcp_server.py runs it after the step is recorded, so the timeline keeps the whole answer."""
|
|
|
|
from .mcp_tools import retext, tool_text
|
|
|
|
# UTF-8 bytes. Antigravity's logs show answers filed from 4,099 bytes (3,627 characters, some of them
|
|
# Chinese at three bytes each) while longer ones were shown at other times: stay under the lowest.
|
|
BUDGET = 3800
|
|
NAME = 60 # characters of a left-out item's name
|
|
|
|
|
|
def size(data):
|
|
return len(tool_text(data).encode("utf-8"))
|
|
# A canvas cut short would read as a smaller canvas: get_situation gives every node in brief.
|
|
WHOLE = {"describe_canvas"}
|
|
|
|
|
|
def _the_list(data):
|
|
"""The key of the one list of items in `data`, or None when there is none or more than one."""
|
|
keys = [key for key, value in data.items() if isinstance(value, list) and value and all(isinstance(item, dict) for item in value)]
|
|
return keys[0] if len(keys) == 1 else None
|
|
|
|
|
|
def fit(result, name):
|
|
"""`result` (an MCP tools/call result) with its one list cut to BUDGET, when it is longer."""
|
|
data = result.get("structuredContent") if isinstance(result, dict) else None
|
|
if name in WHOLE or result.get("isError") or not isinstance(data, dict) or size(data) <= BUDGET:
|
|
return result
|
|
key = _the_list(data)
|
|
if not key:
|
|
return result
|
|
items = list(data[key])
|
|
kept = {**data, key: items}
|
|
while size(kept) > BUDGET and len(items) > 1:
|
|
items.pop()
|
|
left = data[key][len(items):]
|
|
named = [str(item["name"])[:NAME] for item in left if item.get("name")]
|
|
kept["left_out"] = (f"{len(left)} more {key} left out to keep this answer readable in the chat"
|
|
+ (" (more_names)" if named else "") + ": a narrower query, a smaller limit or the next page lists them")
|
|
if named:
|
|
kept["more_names"] = named # what there is, to ask for by name
|
|
if len(items) < len(data[key]):
|
|
retext(result, kept)
|
|
return result
|