claws-memory-enginelisted

# claws-memory-engine — silent orchestration memory layer ## What this is Engine-internal contract governing silent memory recall and write inside four orchestration commands. Memory is infrastructure — the engine pulls context and surfaces excerpts inline. Users never invoke this skill directly. ## When to invoke Fires automatically inside: - `/claws-do` — before the 8-route classifier - `/claws-auto` — before each wave dispatch - `/claws-fix` — after a failure mode is detected - `/claws-plan` — during Step 3, before the worker table ## When NOT to invoke Never invoke directly. No `/claws-memory-engine` slash command exists. For manual memory ops use [[claws-memory]], [[claws-recall]], or [[claws-remember]]. ## Steps **Recall (all commands):** `claws_recall({ query, limit: 5–8 })` before the primary action. On `ok: false` or `CLAWS_NO_MEMORY=1`, skip silently. **Per-command:** - `/claws-do`: query = raw user text; surface ≤3 results inline before classification. - `/claws-auto`: query = wave objective; embed relevant rules as prose in worker missions; log all to `PROGRESS.md`. - `/claws-fix`: query = `<runbook_id> <error_signature>`; include ≤3 matching memories in the diagnosis block. - `/claws-plan`: query = plan goal; cite in Risk Register and Mission Outline. **Remember:** Only after a novel, actionable outcome. Check slugs first — no duplicates. Body: Rule → **Why:** → **How to apply:**. Never auto-write from `/claws-plan`. **Forget:** Never automatically. Ex