rag-evaluatelisted

# rag-evaluate — evaluate index quality with registered probes > Before executing this workflow, read and apply `../HOST_CONTRACT.md`. ## What Automatically throws every pre-registered question in `.noory/rag/probes.json` at the index and reports the **search results + the active AI session's diagnosis** to the user. The deterministic part (search) is handled by the MCP server; interpretation/judgment (where it is weak, what to adjust) is handled by the **active AI session** — the consistent design of this plugin. > This skill is the **measurement stage** before the user hand-tunes the indexing policy. Per the CLAUDE.md `honesty` principle, it does not auto-tune — the user reviews the active AI session's suggestions and applies them directly via `rag_set_settings`. ## Steps 1. **Precondition check**: - Call `rag_get_probes` → if the result is empty: ``` 📭 There are no probes, so nothing can be evaluated. Register 3-5 frequently asked questions first with `/rag:rag-probe-add`. ``` then stop. - Call `rag_stats` → if `chunks == 0`: ``` ⚠️ The index is empty. Run `/rag:rag-reindex` first. ``` then stop. 2. **Batch evaluation**: call `rag_evaluate(k=5)`. A top-5 hybrid search is run at once against every registered probe. 3. **Print the results as a table** (one-line summary per probe): ``` 🔬 Evaluation results (k=5, expand_depth=N) | Probe | Query | Top1 dist | Unique files | Top matches | |-------|-------|