learnlisted

Capture a learning right now, mid-session, without breaking flow. ## Process ### 1. Parse the Insight From the user's input or the current context, determine: - **Type**: Is this a pattern, gotcha, decision, anti-pattern, codebase-fact, or api-behavior? - **Confidence**: Did we verify this (high), observe it (medium), or suspect it (low)? - **Tags**: What files, features, or areas does this relate to? ### 2. Check for Duplicates Read the target `.jsonl` file and check if a similar entry already exists: - If yes → **re-confirm the existing entry** (do NOT append a duplicate): - bump `helpfulCount` by 1 - bump `confidence` if previously `low`/`medium` and the new evidence is strong - set `updatedAt` to now - update `recommendation` only if it's clearly better - If no → append a new entry ### 3. Write Entry Append to the appropriate file in `.agents/knowledge/`. Field order matches `.agents/knowledge/README.md`: ```json { "id": "<type-prefix>-<next-number>", "type": "<detected-type>", "fact": "<concise insight>", "recommendation": "<what to do about it>", "confidence": "<detected-confidence>", "provenance": [{"source": "<human|agent|test-failure>", "reference": "<context>", "date": "<today>"}], "tags": ["<relevant>", "<tags>"], "affectedFiles": ["<paths>"], "createdAt": "<now ISO>", "updatedAt": "<now ISO>", "usageCount": 0, "helpfulCount": 0, "outdatedReports": 0 } ``` `usageCount` is bumped by `/prime` when surfacing, `helpfulCount` b