How it uses Jev
For each ast-grep match, jev-lint sends the relevant code to Jev with a natural-language rule question and receives a probability. The configured cutoff determines whether to report a finding; parsers and matchers run locally.
What Jev decides
The body of this function does something materially different from what its name promises.
Located Go function CountFailed whose body returns failed jobs instead of a count.
App workflow
- Select code matches
Use language-specific ast-grep matchers and configured rules to identify code subjects for review.
- Ask the rule question
Send each matched subject and its rule's natural-language claim to Jev for scoring.
- Report candidates
Compare scores with configured cutoffs, apply suppression and cache behavior, and present findings for human judgment.
Why it is interesting
It applies model judgment to contradictions between code and its own claims, such as comments, names, tests, or failure handling, while documenting uncertainty and supporting cached review workflows.