中文 README

ActionRepro turns GitHub Actions failures into local repro plans and PR evidence packs.

It is not a local runner like act, and it is not a workflow linter like actionlint. It reads the logs you already have, finds the first actionable failure, classifies it, extracts likely local commands, and writes a short report you can use before replying to a maintainer.

pip install actionrepro
gh run view 123456789 --repo owner/repo --log > run.log
actionrepro plan run.log --out repro.md

When a PR check fails, the GitHub UI makes you scroll through a lot of setup noise before the useful line appears. In open-source work, that wastes time and leads to weak replies like "CI failed for me too". A better reply says whether the failure is a real regression, a permission gate, a network limit, a dependency install issue, or a local test failure.

ActionRepro gives you that first pass quickly:

• failure category
• failed job and step
• evidence snippets
• likely local repro command
• PR comment draft

Fetch a run log through gh:

actionrepro fetch owner/repo 123456789 --out run.log

Inspect failures:

actionrepro inspect run.log

Generate Markdown:

actionrepro plan run.log --out repro.md

Generate JSON:

actionrepro plan run.log --format json --out repro.json

Draft a PR comment without posting anything:

actionrepro comment run.log --pr 42 --dry-run

Posting is intentionally disabled in v0.1. The tool should not speak for you or claim a command was run when it was not.

ActionRepro has deterministic rules for common CI failure shapes:

• permission_gate: GitHub token, CLA, Vercel, secret, or integration permission failures
• network_external_service: 429s, failed downloads, or external API/network failures
• runner_disk: ENOSPC and disk pressure
• flaky_timeout: timeout-shaped failures
• dependency_install: package resolution or install failures
• lint_or_typecheck: ruff, flake8, pylint, mypy, pyright, eslint, etc.
• test_failure: pytest/test/assertion failures
• unknown_failure: a real failure marker that needs manual reading

# ActionRepro report

- Log files: 1
- Lines scanned: 37
- Commands detected: 2
- Failures detected: 1

## 1. test_failure

- Job: `test (ubuntu-latest)`
- Step: `Tests`
- Source: `run.log:27`
- Headline: `FAILED tests/test_parser.py::test_parse_windows_log`
- Advice: A test failed and should be reproduced locally.

### Local repro commands

- `python -m pytest -q`

ActionRepro does not simulate GitHub runners. Use act when you need a local runner.

ActionRepro does not lint workflow YAML. Use actionlint for static workflow checks.

ActionRepro does not call an LLM, auto-fix code, or post PR comments. It creates a precise evidence pack so you can make the right call.

python -m pip install -e ".[dev]"
pytest -q
ruff check .
python -m build
twine check dist/*