rag-feedback-reportlisted
Install: claude install-skill noory-code/noory-ai
# rag-feedback-report — feedback aggregation · weak-spot report
> Before executing this workflow, read and apply `../HOST_CONTRACT.md`.
## What
Takes the feedback accumulated via `rag_get_feedback` and the net boost per source (rel_path), and produces a report on **what is searched frequently and which sources match well / poorly**. The **diagnosis stage** of the self-improvement loop — Claude pinpoints weak spots and proposes the next action (source augmentation · reindexing).
## Steps
1. **Aggregation query**: `rag_get_feedback()` → `{feedback:[…], boosts:{rel_path:net}, count}`.
- If empty, guide "No feedback yet. After searching, leave feedback via `/rag:rag-feedback`." then finish.
2. **Aggregation (the active AI session)**:
- **Frequently searched queries**: top query frequencies.
- **Well-matching sources**: rel_paths with the highest positive net boost.
- **Weak-spot sources**: rel_paths with negative net boost — sources that show up in search but the user said were wrong.
3. **Diagnosis · proposal (the active AI session)**: infer why the weak-spot sources don't match (content stale / chunking too large / topic mismatch) + propose an action — source augmentation (`rag-add-source` · `rag-fetch-external`), reindexing (`rag-reindex`), and if contradictory, let the user judge at reindex time.
4. **Output**:
```
📊 Feedback report (N total)
Frequently searched: "…" ×k
Well-matching sources: <rel_path> (+net)
Weak-spot sources: <rel_path>