| name | recall |
| description | Query the cross-provider topic-memory corpus by keyword (and optional class) instead of grepping ~180 files. Use when you need to check "what have we learned about X" before solving a problem. |
| when_to_use | ["recall what we know about a topic","check prior lessons / feedback before implementing","have we hit this before","search memory / topics by keyword"] |
recall — query the topic-memory corpus
One provider-neutral mechanism (Claude / Codex / agy) to query .claude/memory/topics/
with identical, deterministic results. Stdlib-only; no dependencies.
When to use
Before solving a hard or recurring problem, recall prior lessons rather than
re-discovering them. Pairs with .claude/memory/topics/INDEX.md (the browseable index).
Usage
python3 scripts/memory/recall.py "git worktree race" # keyword(s)
python3 scripts/memory/recall.py worktree --class feedback --limit 10
python3 scripts/memory/recall.py --kw stash --kw rebase # repeatable keywords
On Linux with uv: prefix uv run --no-project.
Output is ranked by match count (desc), then filename (stable) — so every provider
sees the same ordered result set. --class filters by problem class
(feedback / project / reference / user / ...).
Related
- Index generator:
scripts/memory/build_topics_index.py → topics/INDEX.md
- Cross-provider memory slices:
scripts/memory/curate_readback_slice.py (codex/gemini/hermes)
- Bridge:
scripts/memory/bridge-hermes-claude.sh (regenerates INDEX + slices)
