COMMUNITY OBSERVEDMetric claims: AUTHOR REPORTEDSOURCE AUDITED🤖 Jev Agent

taste-lint

Catches AI slop in a diff before it ships.

Overview & result

Taste Lint runs design review on every commit, covering motion, copy and typography rules. The mechanical checks need no key or account and can fail the build. Jev review notes are an optional addition that needs a Vercel AI Gateway key. The README says 28 checks block, while 177 more only report, including Jev questions that no labelled corpus has yet promoted.

From the original post

Catch AI slop before you ship.

— Matthew Blode

How Jev fits in the loop

  1. Run taste-lint in your project or from a pre-commit hook.
  2. Mechanical checks for motion, copy, typography, code and search run locally with no key and can fail the build.
  3. Checks marked under review, such as vague errors and empty states with no action, go to Jev, which answers with a probability.
  4. Jev questions report only until a labelled corpus promotes them to blocking.

How to reproduce

  1. Inspect the original showcase and source material at https://github.com/mblode/taste-lint
  2. Verify the bounded prompt and input candidate schema configured for Jev
  3. Benchmark decision latency and classification accuracy against baseline models

Why this build matters

Model-judged checks start as report-only and have to earn the right to block a commit.

Limitations

  • 28 checks block; 177 more only report.
  • The README says no labelled corpus has yet promoted the Jev questions.
  • Jev review notes need a Vercel AI Gateway key and bill it.

Patterns