distill-lessonslisted

# /distill-lessons — surface candidate ledger entries Run after a meaningful chunk of work (a commit, an ADR write, a design discussion, a failed-then-fixed pattern) to extract candidate entries for the append-only lessons ledger. Output goes to `.agent-docs/now/lessons/proposals.md` — **promotion to `.agent-docs/lessons/<slug>.md` is human-gated** (a load-bearing discipline; if you formalize it, record it as an ADR). This skill NEVER writes into `lessons/` directly. **Foreground in the main agent.** Context-capture skills don't background cleanly (same reasoning as `/flush`): the session state they read from is the live conversation. --- ## Recent journal entries (last 50) !`tail -50 .agent-docs/log.md 2>/dev/null || echo "(no log.md yet)"` ## Existing Tier-1 lessons MOC (for the dedup check) @.agent-docs/now/lessons/MOC.md ## Existing proposals (don't duplicate) @.agent-docs/now/lessons/proposals.md --- ## Steps ### 1. Review the inputs above The shell-interpolated `tail -50` gives you the fresh journal. The `@`-imports show what's already in the Tier-1 MOC and staged as a proposal — that is your dedup baseline. If a candidate is already covered by a MOC row or an existing proposal, skip it. ### 2. Scan the session for the five candidate TYPES Walk the recent work through each lens (the `type:` axis of `templates/lesson-template.md`). Most sessions surface zero to a few — that is fine: - **false-belief** — you (or an agent) acted on something that turned out