Model Setup Maintainer

Use this skill for local and cloud vision model readiness.

Evals

Should trigger

• "update the YOLO weights"
• "switch the Ollama vision backend"
• "doctor model readiness check fails"
• "document the OpenAI vision setup"

Should not trigger

• "change MCP tools" → use mcp-tools-maintainer
• "change CLI flags" → use cli-maintainer
• "docs only" → use docs-maintainer

Core Files

• docs/model-setup.md
• mq_image_analyze/cli/doctor.py
• mq_image_analyze/vision/detection/detector.py
• mq_image_analyze/vision/semantic/provider.py
• mq_image_analyze/vision/semantic/ollama_vision.py
• mq_image_analyze/vision/semantic/openai_vision.py
• tests/test_doctor.py
• tests/test_semantic_provider.py

Model Contracts

Detection:

• YOLOv8n weights are expected locally.
• Model weights must not be committed.

Semantic vision modes:

• local-fast defaults to bakllava via Ollama.
• local-deep defaults to llama3.2-vision via Ollama.
• cloud-verify defaults to gpt-4.1 via OpenAI.
• Use gpt-4o when specifically requested; do not invent gpt-4.0.

Change Rules

• Keep doctor output clear about missing dependencies and model files.
• Keep model names consistent across README, CLI docs, and provider code.
• Do not require cloud credentials for local modes.
• Do not download or commit model weights automatically without explicit user request.

Verification

python -m pytest tests/test_doctor.py tests/test_semantic_provider.py -q
mq-image doctor --json

For docs changes:

rg "local-fast|local-deep|cloud-verify|gpt-4.1|gpt-4o|bakllava" README.md docs mq_image_analyze