knowledge-updatelisted

# Knowledge Update Extract what was learned in this session and persist it to the project's domain memory files. ## When to Use Run at the end of a session with `/knowledge-update` when: - You just completed a non-trivial task - You encountered a surprising bug, API quirk, or pattern - Something worked differently than expected - A previous hypothesis was confirmed or contradicted ## How It Works ### Step 1 — Identify the domain Look at what was worked on. Map it to an existing domain file or decide if a new one is needed. Existing domain files live in the project memory folder: `~/.claude/projects/<project-slug>/memory/domain_*.md` Only create a new domain file if the work clearly belongs to a different domain than existing ones. ### Step 2 — Extract insights from the session Review the conversation and identify: **Facts** (certain, repeatable, no ambiguity): - API behaviors, schema details, confirmed constraints - "Property names are case-sensitive" — not "I think they might be" **Hypotheses** (pattern seen, not confirmed enough): - Things that worked once or twice but need more evidence - Add to the hypotheses table with confirmation count 1 **Confirmations** (a hypothesis proved true again): - Find the matching row in the hypotheses table - Increment the confirmation count **Contradictions** (a rule proved wrong): - Move it back to Hypotheses with a note ### Step 3 — Apply promotion logic - Hypothesis with **5+ confirmations** → move to Rules section - Rule