Skill: recall

Trigger phrases:

• /recall [question or topic]
• "what did we decide about X"
• "do we know anything about X"
• "have we seen X before"
• "what's in memory about X"
• Any quick single-topic lookup

Use search-memory instead when: the question could span 3+ memory categories and you want parallel agents with a synthesized answer.

Steps

Step 1 — Extract the query

Pull the search term(s) from the user's message. If it's a full question ("why did we move files to mobile/?"), extract the key nouns: "mobile files location".

Step 2 — Run the search

Try semantic search first (finds related memories even with different wording). Falls back to keyword search automatically if the index doesn't exist or sentence-transformers isn't installed.

python tools/memory.py --search-semantic "[query]" --top 8

Run this via Bash. If semantic search returns no matches, also run keyword search:

python tools/memory.py --search "[query]" --top 8

Step 3 — Present results

If results found:

## Recall: "[query]"

**[file name]** (score N)
> [matching lines]

**[file name]** (score N)
> [matching lines]

---
Bottom line: [1 sentence summary of what memory says]

If nothing found:

No memory found for "[query]".

This topic hasn't been captured yet. If it comes up this session, run /learn at the end to save it.

Step 4 — Offer to dive deeper

If a result looks relevant but partial:

"Found in lessons.md — want me to read the full entry?"

Notes

• Never write or modify memory files in this skill — read only
• If memory.py is not found, fall back to: Grep(pattern="[query]", path="[MEM]") recursively
• This skill is intentionally fast — no agents, no synthesis overhead
• For broader research across multiple categories, use search-memory