recall

name	recall
description	Search Memex long-term memory. Returns facts, notes, and entities matching the query.
argument-hint	[search query]

/recall — Search Memex

1. Query source:

   • If $ARGUMENTS is non-empty, use it as the search query.
   • If $ARGUMENTS is empty, look for a "Composed query" block in the additional context — the plugin's UserPromptExpansion hook supplies one when you invoke /recall without arguments. Print a one-line summary of the composed query so the user can see what's being matched.
   • If neither is available, ask the user.

2. Search strategy:

   • Vault scope (applies to every search/list call below): if the SessionStart "Per-project vault" block names a vault, pass it — memex_memory_search/memex_note_search/memex_list_entities scope via vault_ids=["<that vault>"]. If no project vault is set, omit vault_ids (server default). To search every vault, pass vault_ids=["*"]. (Listing without it queries the server's default read scope, which is usually NOT the project vault — the cause of "nothing found" when the briefing shows the vault has content.)
   • How-to queries first ("how do I X?", "how do we deploy?", "what's the checklist for Y?"): call memex_procedural_search ONLY (see "Procedure recall" below). Do NOT also run memex_memory_search / memex_note_search for a how-to — the procedural plane is the only home for how-tos.
   • Otherwise (facts, decisions, context): memex_memory_search AND memex_note_search in parallel (both scoped per the rule above). If insufficient, retry with expand_query=true.
   • If still nothing, try memex_list_entities. If all fail, say so — do not guess.
   • Historical queries ("how has X evolved?", "show me everything/hidden"): use memex_get_unit_history(unit_id) or memex_memory_search(apply_pre_filter=False).

3. Present results: summarize clearly, include source Note IDs.

4. Procedure recall: for "how do I X?" queries, the procedural plane is the ONLY source — there is no KV fallback for how-tos.

   • Pinned cards: the highest-value procedures are already in your SessionStart briefing (pin chain global → project:<id> → app:claude-code) — answer from there before searching.
   • Procedural search: memex_procedural_search(query="...", kind="procedure") for hybrid BM25+vector search; memex_procedural_get_by_identity(kind="procedure", scope="global", verb="...", context="...") for an exact anchor lookup. The plane carries procedure / strategy kinds under the identity anchor (kind, scope, verb, context); strategies anchor on (scope, verb) with no context.

5. Memory hygiene: when asked about stale facts, call memex_get_lint_flags (omit vault_id to use the active vault, or pass the project vault). Each finding's id is the id field; rule_name discriminates the type. Surface findings for the user to act on — do NOT apply autonomously: a canned resolve --action / memex_lint_apply_winner is server-gated and returns HTTP 403 unless the server runs with MEMEX_LINT_ALLOW_UNATTENDED_APPLY=1. For the full resolve/dismiss workflow use /lint.

6. Diagnostics: "how is this vault doing?" → memex_get_diagnostics_summary (omit vault_id for the active vault, or pass the project vault) — returns cluster count, top entities, and MW score distribution.

Transcript-aware fallback contract

When you invoke /recall with no arguments, a UserPromptExpansion hook reads the conversation transcript, extracts the last MEMEX_CC_RECALL_TURNS turns (default 3, range 1–10), and injects a composed query as additional context. The composed text is enclosed between --- Composed query (last N turns) --- markers so the skill can locate it deterministically. If an explicit query is supplied, the hook does not run — your $ARGUMENTS always wins.