[Prateek99-s]

Description:

## Summary of changes

- Added `outputs.is_formatted` to `action.yml` and attached `id: black` to the composite execution step.
- Updated `action/main.py` to write `is_formatted=true` / `is_formatted=false` to `$GITHUB_OUTPUT` based on Black's exit code

and report output.
- Documented the new output in docs/integrations/github_actions.md.
- Closes #3526.

## Test Plan

- Ran `mypy action/main.py` (clean, 0 errors).
- Ran `pytest tests/test_docs.py` (passed).

[sylvesterkaczmarek]

--quiet seems to make this output wrong in normal write mode. Black can reformat files, return 0, and suppress the reformatted summary, which leaves is_formatted=false. Since the action lets users pass --quiet, could we derive this without parsing human stdout, or at least cover quiet mode in a regression test?

[Prateek99-s]

Good catch! When --quiet was passed, Black's stdout summary was completely silenced in write mode, causing is_formatted to default to false.

To resolve this reliably without needing external filesystem diffing:

1. We strip -q/--quiet from the arguments passed to the internal Black subprocess so Black always produces its standard return code and summary internally.
2. is_formatted is derived accurately using returncode == 1 or checking for reformatted file counts via regex.
3. If --quiet was requested by the user, we filter the output prior to printing/writing to OUTPUT_FILE (suppressing summary banners while preserving diff output and error diagnostics).
4. Added a regression test suite in tests/test_action.py covering write, check, diff, quiet, and error scenarios.

[cobaltt7]

See #3531 - the preferred implementation would not be reading stdout/stderr which is meant for humans, not machines. See POC commit rasjani@367beb8

[sylvesterkaczmarek]

Agreed. Using a machine-readable signal rather than parsing Black's human stdout/stderr is the cleaner direction and also removes the need to strip/replay --quiet just to infer the result. That addresses the concern I raised more robustly; I'd follow the #3531/POC approach from here.

[github-actions]

diff-shades results comparing this PR (66e13c7) to main (b5b1eb6):

--preview style: no changes

--stable style: no changes

---

What is this? | Workflow run | diff-shades documentation

[Prateek99-s]

Thanks for pointing me to #3531! I've refactored this to follow the POC approach: Report.write_github_outputs() now write directly to $GITHUB_OUTPUT when present, based on Black's internal change_count. This completely removes the need for stdout parsing, regex matching, and flag stripping in action/main.py. I also added tests in tests/test_black.py covering normal, quiet, and check modes. Ready for review!

[sylvesterkaczmarek]

Rechecked current 92a7ffa. The implementation now derives the GitHub Action output directly from Black's internal Report.change_count instead of parsing stdout/stderr, so --quiet no longer hides the signal. The CLI regressions cover normal, quiet and --check modes, and the current CI matrix is green. This resolves the issue I raised.

[sylvesterkaczmarek]

Rechecked current 1c5ee61. The action still derives its machine-readable result from Report state rather than human stdout/stderr, so the --quiet issue remains resolved. The current head also exposes change_count/same_count/failure_count through the composite step, documents them, and keeps regression coverage for normal, quiet and --check modes. The full current CI matrix is green. No blocker from me.

[cobaltt7]

Tested and confirmed this works in another repo! Thanks!