recall-codebase

Recall codebase memory (read, don't re-explore)

The whole point of explorer is: explore once, then read the memory forever. Before any agent re-scans the code, it must check whether a fresh memory already answers the question.

Procedure

1. Check for .claude/explorer/MEMORY.md. If absent, there is no memory — a full exploration is needed (run /explorer:start).
2. If present, read MEMORY.md fully. It is the authoritative summary (what / why / how).
3. Read the explored_commit: field at the top. Compare to current git rev-parse HEAD:
   • Equal → memory is current. Answer from memory + map/<area>.md + index.json. Do not re-explore.
   • Different → run git diff --name-only <explored_commit> HEAD to see what changed. If changes are unrelated to the question, still answer from memory but flag it. If they touch the relevant area, do a targeted re-read of just those files (not a full scan), or recommend /explorer:start for an incremental refresh.
4. To answer a specific question, use index.json to jump straight to the relevant files and their map/<area>.md deep-dive instead of reading the whole tree.
5. Trust unverified: markers: if the answer lives in an unverified area, say so rather than guessing.

Output contract

When you recall, tell the user (briefly): whether memory was current, which files you used, and any staleness caveat. This keeps the "understand by reading" path trustworthy.