Score code quality for coding agents

01 · Role in the system

What Jev does here

For each file Jev answers a fixed set of Noul questions such as long method, deep nesting, or complex conditional, and the CLI combines those probabilities into per-file and repository scores where every component traces to a checkable claim. Answers are cached by content hash so reruns pay only for changed files, while test coverage runs, assertion mapping, and gap queries stay entirely local and free.

02 · Control boundary

Where Jev sits

Per-file structural facts to batched yes-or-no judgments; deterministic arithmetic ranks files and feeds an agent loop that proves each fix with coverage diffs.

Code owns the loop, permissions, thresholds, validation, and side effects. Jev owns only the bounded judgments described above.

03 · Known limits

What this evidence does not prove

• Quality scores are model judgments over fixed structural questions, not measured defect rates.
• The cost claim assumes published pricing and excludes any teacher or fallback models an agent adds around the tool.

04 · Attribution

Public sources

This is a Community record: the project was published by a third-party community author.

• supercorp-ai ↗Community · github · public · checked 2026-09-18