knowledge-looplisted

# Knowledge Loop Unifies the three knowledge systems (RAG index, claude-mem, handoffs) into one workflow so capture and retrieval stop being separate manual acts. ## Auto-invocation triggers - User asks "what did we decide about X" / "where did we leave Y" / "is there a memory note for Z" - End of a meaningful task (commit landed, PR merged, decision reached) - User explicitly says "remember", "save this", "checkpoint", "handoff" - Session-budget guard signals approaching context limit ## Workflow **Mount guard (required before any RAG/brain op — `standards/skill-patterns.md §mount-guard`):** run [references/mount-guard.sh](references/mount-guard.sh) — if External HD is unmounted, surface `BLOCKED: External HD unmounted — RAG/vault unreachable` and halt; do not return empty recall as if the index were searched. ### Phase 1 — Query (always) Invoke `/recall` (MCP: `rag_query(query="<topic>", top=5)`) with the user's question or the active task topic. For which knowledge source to route to, `recall` is canonical — see [references/recall-routing.md](references/recall-routing.md). If the user is asking a recall question, return the answer immediately and skip Phase 2/3 unless they also asked to capture something. **Done when:** recall returns `{hit_count: N, top_cosine: X, source: [memory|handoff|commit|code]}` — confirm top result answers your question (cosine ≥0.50) or increase `top` up to 8. ### Phase 2 — Capture (if new knowledge produced) Invoke `/sync-memories` with