fix(hook): do not stage fixes when fail_on_fix=true

[jdx]

Summary

When a hook has fail_on_fix=true, the step's auto-staging would silently re-stage the fixer's output over the user's explicit git add choices. After the failed commit:

• The user's staged changes were merged with / overwritten by the fix
• The fix was no longer visible as an unstaged change for review
• A re-commit would silently succeed with the fix baked in — defeating the entire point of fail_on_fix

This is the bug reported in #888.

Fix

Force should_stage = false during fix runs when fail_on_fix = true (src/hook.rs:828). The fixer's output stays in the worktree as an unstaged change, the user's staged changes are preserved, and a re-commit will fail again until the user explicitly accepts the fix.

After the fix, the user's reproduction scenario yields exactly what they expected:

=== git status AFTER failed commit ===
Changes to be committed:
        modified:   a.json        # user's original staged change preserved
Changes not staged for commit:
        modified:   a.json        # fixer's reformatting now visible for review
        modified:   b.md          # untouched

Test plan

• New regression test in test/fail_on_fix.bats reproducing the discussion scenario
• All 5 fail_on_fix.bats tests pass in both libgit2 and nolibgit2 modes
• All other stage / stash bats tests still pass
• cargo test (145 tests) passes
• Manual reproduction matches expected behavior

Closes #888

🤖 Generated with Claude Code

---

Note

Medium Risk
Changes pre-commit/hook execution behavior around git staging during fix runs, which can affect users’ commit contents. Scope is small and covered by a new regression test, but it touches git index/worktree semantics.

Overview
Ensures fail_on_fix=true runs do not auto-stage fixer modifications by forcing should_stage off for RunType::Fix, so a failed hook leaves fixer edits as unstaged changes for review and does not overwrite the user’s staged selection.

Adds a Bats regression test reproducing #888 to verify a failed pre-commit preserves the original staged diff while exposing the fixer’s reformatting in the working tree.

^{Reviewed by Cursor Bugbot for commit 7a2b202. Bugbot is set up for automated code reviews on this repo. Configure here.}

[greptile-apps]

Greptile Summary

This PR fixes a bug (#888) where fail_on_fix=true hooks would silently auto-stage the fixer's output, overwriting the user's deliberate git add choices and allowing a re-commit to succeed without the user explicitly reviewing the fix. The fix is a single targeted condition in hook.rs that forces should_stage = false when both fail_on_fix is set and the run type is Fix, keeping the fixer's output as an unstaged diff for review.

Confidence Score: 5/5

This PR is safe to merge — it is a minimal, well-targeted fix with a solid regression test covering the exact bug scenario.

The change is two lines touching a single, narrow condition. The logic is correct: forcing should_stage=false when fail_on_fix=true && run_type==Fix has no unintended side effects. The bats test faithfully reproduces the original bug report and verifies both sides of the invariant. No security or data-integrity risks introduced.

No files require special attention.

Important Files Changed

Filename	Overview
src/hook.rs	Two-line fix: overrides should_stage to false when fail_on_fix=true and run_type=Fix, correctly preventing fixer output from being staged and defeating the purpose of fail_on_fix.
test/fail_on_fix.bats	New regression test reproducing the exact issue from #888: verifies staged changes are preserved and the fixer's output remains as an unstaged diff after a failed commit.

Sequence Diagram

sequenceDiagram
    participant User
    participant Git
    participant HK as hk (pre-commit)
    participant Fixer

    User->>Git: git add a.json
    User->>Git: git commit -m "update"
    Git->>HK: run pre-commit hook
    Note over HK: fail_on_fix=true, run_type=Fix<br/>→ should_stage forced to false
    HK->>Fixer: run fix step on a.json
    Fixer-->>HK: a.json modified (spaces stripped)
    Note over HK: should_stage=false → skip git add
    HK-->>Git: fail (files were modified by fixer)
    Git-->>User: commit rejected

    Note over User,Git: After fix (correct behavior)
    User->>Git: git diff --cached a.json
    Git-->>User: shows user's original staged change (a.json with spaces)
    User->>Git: git diff a.json
    Git-->>User: shows fixer's unstaged change (spaces stripped)

Loading

_{Reviews (1): Last reviewed commit: "fix(hook): do not stage fixes when fail_..." | Re-trigger Greptile}

[gemini-code-assist]

Code Review

This pull request modifies the hook execution logic in src/hook.rs to prevent automatic staging of fixes when fail_on_fix is enabled, ensuring that fixes are surfaced for manual review. Additionally, a new integration test in test/fail_on_fix.bats verifies that existing staged changes are preserved and fixes appear as unstaged changes. I have no feedback to provide.