← ClaudeAtlas

recalllisted

Run explicit Signet recall through the canonical scoped recall path, preserving scores, sources, provenance, supplementary context, and session dedupe metadata.
Signet-AI/signetai · ★ 288 · AI & Automation · score 81
Install: claude install-skill Signet-AI/signetai
# /recall Use this skill for targeted explicit recall. It is not the same surface as prompt-submit context injection. Signet's canonical explicit recall endpoint is `POST /api/memory/recall`. The CLI, MCP, and hook recall surfaces should stay thin wrappers around that contract. Recall combines FTS5, prospective hints, vector similarity, structured path evidence, graph traversal, optional reranking, source-backed fallbacks, currentness shaping, and session context dedupe where configured. Do not describe it as a fixed 70/30 vector/BM25 search. ## When To Use Use `/recall` when: - the user asks what Signet remembers or asks for a targeted memory search - current context is missing an old decision, preference, project fact, or prior source - you need provenance, source labels, scores, ids, or no-hit metadata - you need aggregate recall to synthesize a bounded answer from evidence - you are debugging recall quality, scoping, or session-dedupe behavior Do not use `/recall` as a ritual before every task. Session-start and prompt-submit injection already provide lightweight context. Use explicit recall when there is a concrete retrieval question. ## CLI ```bash signet recall "<query>" ``` Useful options: ```bash signet recall "Signet ontology policy" --agent codex --limit 10 signet recall "vim keybindings" --type preference --tags editor signet recall "OpenMarketUI evaluator health" --project /mnt/work/openmarketui signet recall "what did we decide about source truth" --