prompt-tunelisted

# Prompt Tuning Evaluate and tune Myco's LLM prompt templates against local models. Run structured test fixtures, compare model performance, identify prompt weaknesses, and iterate until the prompt passes reliably. This is a development tool for the Myco project — not shipped to end users. ## Prompt Inventory All prompt templates live in `src/prompts/`. Current status: | Prompt | Purpose | Has Fixtures | Tuned | |--------|---------|:---:|:---:| | `consolidation.md` | Consolidate related spores into wisdom notes | Yes | Yes | | `supersession.md` | Detect when a new spore supersedes an old one | No | No | | `extraction.md` | Extract observations from session events | No | No | | `summary.md` | Generate session summaries | No | No | | `title.md` | Generate session titles | No | No | | `classification.md` | Classify artifacts | No | No | | `session-similarity.md` | Detect parent-child session relationships | No | No | | `digest-*.md` | Synthesize vault knowledge into tiered extracts | No | No | ## Workflow ### When fixtures exist: Run the eval 1. Check which prompts have eval fixtures: `ls tests/prompts/*-fixtures/ 2>/dev/null` 2. List available models: `curl -s http://localhost:11434/api/tags | python3 -c "import json,sys; [print(m['name'], m['details']['parameter_size']) for m in sorted(json.load(sys.stdin)['models'], key=lambda x: x['name'])]"` 3. Run the eval for a specific prompt: ```bash # Single model EVAL_LLM=true EVAL_MODEL="phi4:latest" npx vitest run tests/pro