Overview & result
Jev Review Action reads PR patches or catalog submissions, asks Jev policy questions and a category Choice, then writes a fixed-template advisory comment. It does not label, approve, merge, or edit the PR.
From the original post
Review submissions and classify pull requests with TypeSafe Jev. You define the criteria and categories; Jev returns typed judgments; code applies the policy and updates one PR comment.
Only Jev is used. Comments come from a fixed template. There is no text-generation model, autonomous agent, or hosted bot server.
How Jev fits in the loop
- Run a privileged workflow from trusted base-branch code and collect PR or catalog evidence.
- Ask Jev positive yes/no policy checks and one bounded category question.
- Apply deterministic policy to return recommended, needs-review, not-recommended, error, or skipped.
- Update a fixed-template PR comment and report artifact without automatically approving or merging.
How to reproduce
- Inspect the source implementation at https://github.com/fatwang2/jev-review-action.
- Configure local Jev API credentials and register required tool/model bindings.
- Execute test prompts and observe decision latency and routing outputs.
Why this build matters
Shows how typed Jev decisions can inform a repository's review workflow while human approval and GitHub actions remain separate.
Limitations
- An independent personal project, not an official TypeSafe product.
- Comments come from a fixed template; no text-generation model is involved.
- Ambiguous answers, category disagreement or missing evidence produce needs-review instead of a verdict.
- The privileged workflow must never check out or execute the PR head.