remem-mcplisted

You have a long-term memory server via MCP. Use the tools automatically — do not ask permission. ## Tools **Core:** `recall` `capture` `search` `feedback` `forget` `resolve` `handoff` `adr` `update` `consolidate` `scenario_create` `persona_update` **CodeGraph:** `codegraph_index` (run once per session; idempotent) `codegraph_search` `codegraph_callers` `codegraph_callees` `codegraph_impact` `codegraph_list` `codegraph_stats` `codegraph_detect_changes` (git diff → affected symbols) **Wiki:** `wiki_ingest` `wiki_search` `wiki_get` `wiki_outdated` **Canvas (F1):** `canvas_get` (Mermaid graph for session) `ref_read` (drill down to raw tool output by node_id) **Skills (F3):** `skill_create` (manual skill with auto-versioning) `skill_archive` (force always-inject) `skill_get` `skill_list` `skill_search` ## Rule 1: Recall before answering Hooks auto-inject BM25-only memory (shallow). **MUST call `recall()` at the start of every non-trivial task** — it does hybrid search (BM25 + vector) with more results and filters. ``` recall({ "query": "<user's question or task>", "mode": "hybrid", "limit": 5 }) ``` After using recall results, call `feedback` for the most useful result. This creates a flywheel — useful memories rise, unhelpful ones fade. ``` feedback({ "capture_id": "<id from recall result>", "signal": "helpful" }) ``` **Signals:** `helpful` (answered question) · `not_helpful` (irrelevant) · `stale` (outdated) · `wrong` (incorrect) ## Rule 2: Capture after non-trivial wor