Check code against team rules written in plain English
Write house rules that linters cannot express and have Jev check each changed file against them on every pull request.
Author's figure: Author reports both rulesets on four files cost about seven cents, and batching comments with file context cut false flags from 25 to 2.. Reported by the builder, not measured by us.
The use case
Teams have conventions that linters and type checkers cannot express, such as never logging customer emails or always going through the billing service. This tool lets the team write those rules as ordinary sentences; Jev then tests each changed file against each rule, locally or in CI. A broken rule is reported with the file so the author can fix it before review. The rules file lives in the repository and is reviewed like code. Reviewers still read the change; the check catches known mistakes, not new ones.
Example workflow
- Write rules in a plain-English rules file in the repo
- Run the checker locally or in GitHub Actions on changed files
- Ask Jev a Noul question per rule per file
- Post broken rules as PR comments
- A reviewer confirms and the author fixes
Tools
- Jev (TypeSafe AI)
- GitHub Actions
- Claude Code
Risks and guardrails
- Vague rules produce noisy comments
- Large files may exceed useful context
- False confidence if rules are treated as full review
Sources
Written by Agents Lookup. Credits and official pages for the tools this use case names:
- Hunch by Kelbie on Made with Jev: madewithjev.com/builds/hunch
- Hunch: original post or repo by Kelbie: github.com/Kelbie/hunch
- GitHub repo by Selmar: calibrating Jev as a code reviewer: github.com/Selmar/typesafe-jev-calibrate-for-code-review
- Hacker News submission: news.ycombinator.com/item