fix(aios-commit): --vault commits a staged rename as a rename

[matiasmacera]

What

aios-commit --vault could commit a rename as a copy. When the rename was already staged in the real index, both names ended up in HEAD. That happens with git mv, or with an editor that stages its own moves.

The cause is that the sweep lists paths with git diff --name-only HEAD, and rename detection is on by default for git diff. A staged rename is therefore reported only under its new name. The commit is built in a throwaway index seeded from HEAD, so it added the new path and kept the old one. A plain mv with nothing staged was unaffected, because it shows as a deletion plus an untracked file, and that is why this hid.

Fix

--no-renames on both of the sweep's diff calls: the one against HEAD, and the --cached one for a repo with no commits yet. Both sides of a rename are then listed. The pathspec is unchanged, so a move into or out of vault/ contributes only the side inside it.

A vault that already holds a duplicate heals itself. The old file is no longer on disk, so the next --vault run reports it as a deletion.

Proof

tests/aios-commit.test.sh, which already runs in both CI lanes, gains four cases:

• git mv then --vault: HEAD holds only the new name.
• Plain mv then --vault: HEAD holds only the new name, the control.
• The same git mv against the hook pinned at the pre-change commit: both names land in HEAD. This case is skipped with a message when that commit is absent.
• A duplicate made by the old hook, followed by one run of the new one: the old name is removed.

Removing --no-renames fails the first case. tests/commit-scan-fail-closed.test.sh fails one sweep call through a git shim matched on the exact argv, so the shim now matches the new argv too. 34 pass under bash 5 and bash 3.2, and the full tests/ run passes.

Scope

hooks/aios-commit (two flags and a comment), the two tests, and the CHANGELOG entry. It closes the last aios-commit item from the hooks audit in the series announced in #158.

🤖 Generated with Claude Code

[chuycepeda]

Thank you, Matías. This shipped in v0.8.6 (merged via #195). One follow-up on top: after a staged rename, the post-commit index sync tried to add the old path, which no longer exists, and printed a false warning. It now skips paths that are gone from both the disk and the index, and the test checks that no warning appears and the status is clean.