learnlisted

@markdownai v1.0 # /learn — Extract and Edge Use when something non-obvious was discovered: a workaround, a subtle invariant, a constraint that affects future work. Learned patterns live as edges in the knowledge graph, not as orphaned files. They are discovered by traversing relationships, not by searching. ## What to learn - Operational constraints discovered during work - Workarounds for library/API quirks - Version-specific fixes or incompatibilities - Architecture patterns or gotchas - Integration patterns that save time - Performance characteristics not obvious from code ## What NOT to learn - Code patterns derivable by reading the repo - Git history (use `git log`) - Task state or in-progress work from this session - Anything already in CLAUDE.md or documented code ## Steps 1. **Name the pattern** — one short title. Examples: - "WillowStore auto-embeds on __init__ when _VEC_AVAILABLE" - "Skills consolidation by mode reduces duplication" - "Gleipnir rate window resets per app_id independently" 2. **Search existing atoms** — `kb_search` for the pattern name and related keywords. If an identical pattern atom exists, update its edges instead of creating a duplicate. 3. **Identify related atoms** — list 2-4 atoms this pattern connects to: - Problem it solves - Module/system it affects - Similar patterns or constraints - Use cases that trigger it 4. **Create the pattern atom** — `kb_ingest`: ```json { "app_id": "hanuman", "tit