Overview & result
An open-source GitHub PR labeler uses Jev's typed decisions to choose labels from a fixed taxonomy and judge size by conceptual scope, not line count. A CLI supports dry runs, and an optional GitHub Action applies labels after review.
From the original post
Size means conceptual scope, not lines changed.
How Jev fits in the loop
- Read a pull request and reviewed current-head context.
- Ask Jev bounded questions for primary type, areas, special attention and conceptual size.
- Report confidence; optionally apply allowed taxonomy labels to GitHub.
How to reproduce
- Clone the repository, configure a spend-capped OpenRouter key and GitHub authentication.
- Run `python3 -m jev_labeler --repo owner/repo --pr 123` for a dry run; inspect confidence before `--apply`.
- Review the example GitHub Action and pin it to a full commit SHA before deployment.
Why this build matters
Jev classifies a PR while GitHub automation and humans retain control over repository changes.
Limitations
- A dry run still makes a paid Jev call; only `--apply` writes labels.
- The documented workflow waits for a current-head Greptile review and abstains on low-confidence categories.