BACK-471 - Fix cli-commit-behaviour tests using git rev-list --count HEAD

[lenucksi]

Basic problem here: If you use a tool that also writes to separate refs per commit, this will break a bunch of tests.

Namely these here:

✗ CLI Auto-Commit Behavior with autoCommit: true > should commit when creating a task if autoCommit is true [608.01ms]
✗ CLI Auto-Commit Behavior with autoCommit: true > should commit when creating a document if autoCommit is true [574.01ms]
✗ CLI Auto-Commit Behavior with autoCommit: true > should commit when creating a decision if autoCommit is true [581.01ms]

This ensures they're not breaking in such contexts.

Summary

• getCommitCountInTest in src/test/cli-commit-behaviour.test.ts used git rev-list --all --count, which includes commits in refs/notes/ai — a ref written asynchronously by a global git-ai daemon configured via trace2.eventTarget in ~/.gitconfig.
• Because the daemon writes note commits after a short delay, the count taken before the CLI command (initialCommitCount) and after it (finalCommitCount) could differ by 1 even when no real commit was made, causing all 6 tests to fail non-deterministically.
• The fix is a one-character change: --all → HEAD. git rev-list --count HEAD counts only commits reachable from the current branch tip, which is the correct metric for "did this CLI command produce a commit on this branch?". This matches the approach already used by the sibling tests in cli.test.ts.
• No implementation changes are needed; shouldAutoCommit and all three CLI handlers (task create, doc create, decision create) already read autoCommit from config correctly.

Test plan

• bun test src/test/cli-commit-behaviour.test.ts — 6/6 pass (was 0/6)
• bun test — 1256 pass, 0 fail, no regressions
• bunx tsc --noEmit — clean
• bun run check . — clean

🤖 Generated with Claude Code

[MrLesk]

Alex's Agent: Thanks, the git rev-list --count HEAD change itself looks reasonable for this test. Counting only commits reachable from the current branch is the right signal for these auto-commit assertions, and it avoids unrelated refs such as notes.

I do not think this PR is ready to merge as-is though, because the branch includes unrelated terminal-status configuration changes in addition to the BACK-471 test fix.

Could you please split or rebase this PR so it only contains the src/test/cli-commit-behaviour.test.ts change?

Specific blockers:

1. The PR title/body describe a one-line test fix and say no implementation changes are needed, but the branch also changes src/file-system/operations.ts, src/types/index.ts, src/utils/terminal-status.ts, and src/test/terminal-status.test.ts for terminalStatuses.

2. The included terminalStatuses config work is incomplete. It parses terminal_statuses, but serializeConfig does not write it back, so a later config save can silently drop that setting.

3. The new helper accepts a terminalStatuses override, but the actual runtime call sites still pass only statuses: CLI cleanup, core cleanup selection, the Web board cleanup affordance, and the All Tasks terminal-status filtering. So the tests cover the helper directly, but not behavior users would actually see.

4. I do not see a matching Backlog task for BACK-471 in this branch. If this PR is kept, it should include the task details according to the project workflow.

I would be happy with the HEAD test-count change as a narrow PR once the unrelated terminal-status changes are removed or moved into a separate, complete task/PR.

[lenucksi]

Branch has been force-pushed and cleaned up. The PR now contains exactly one commit with two files:

• src/test/cli-commit-behaviour.test.ts — the single-line fix (--all → HEAD)
• backlog/tasks/back-471 - Fix-CLI-auto-commit-for-doc-decision-task-create-commands.md — task record (status: In Review)

All unrelated terminal-status changes from BACK-466/467 have been removed. bun test src/test/cli-commit-behaviour.test.ts passes (6/6).