shrug-prompter 0.6.0 — rewrite as heylookitsanllm client
Fresh history. 10 ComfyUI V3 nodes (io.ComfyNode + ComfyExtension + async
execute) in a single nodes.py. Async ShrugClient in client.py covers every
heylookitsanllm inference endpoint: chat, chat_multipart, chat_batch,
embed, transcribe, models, capabilities, cache_list, cache_clear.
- client.py async httpx client, persistent AsyncClient per instance
- media.py tensor <-> JPEG/PNG/WAV bytes
- nodes.py 10 nodes (Connection, VLM, VLMBatch, Template, Text
Cleanup, Accumulator, ASR, Embeddings, Frame Pair,
Cache Control) + frontmatter parser
- __init__.py V3 probe + /shrug/get_models aiohttp route
- web/provider.js model-dropdown widget
- scripts/smoke.py live LAN exerciser
- templates/ frontmatter-aware .md library + 250 style prompts
- tests/ 67 tests (client mocked via respx, media, nodes, templates)
- .claude/ shrug-test / shrug-node-add / heylook-endpoint-sync skills,
vlm-workflow-reviewer agent, PreToolUse block on uv.lock +
templates/styles edits
Fred Bliss
99ad93e21c