[kaanozhan]

During a commit on 2026-10-01, the pre-commit hook (update-structure.js --staged) blocked the commit for 170 s:

• git hash-object --stdin, called by STR-02c's mirror step to hash the tracked map on disk, sat in read() waiting for stdin to close, while node waited inside spawnSync.
• Killing the stuck subprocess let the commit finish, with "mirror": "failed" recorded.
• The same command outside a commit (0.3 s) and a 1 MB-map commit in a temporary repository (0.4 s) did not reproduce it, so the root cause looks environmental. A commit hook must never hang the commit either way.

Changes (scripts/structure-commit.js)

• blobId(bytes, like): Git's blob id computed in process (sha1("blob <len>\0" + bytes), or sha256 when compared with a 64-hex id). It replaces the three hash-object --stdin calls that piped the whole map through Git: the disk-vs-staged check in mirrorToDisk, repairPathspecIndex, and reconcileTrackedMap. No filters applied to --stdin before either, so the result is identical.
• Every git() call has a 60 s limit (SIGKILL). A stuck call fails the run with reason git-timeout, the hook's || true lets the commit through, and the receipt records it, so check-freshness reports that the commit kept the previous map.

Tests

• blobId equals git hash-object for empty, JSON and binary/UTF-8 bytes, and follows a SHA-256 id.
• A fake git that sleeps fails within the limit with git-timeout.
• npm test: 1111 passing.