owned
gauntlet
A local tool and Claude Code plugin that runs a Python project's tests, measures how good those tests really are, and fails when the code is not proven, so you read evidence instead of an agent's diff.
Highlights
- Gates on CRAP score (limit 8) and mutation score (90% of planted bugs caught), with
mutmutorcosmic-ray. - Built for an untrusted author: limits live outside the repo, gates fail closed, and results are hashed against the source.
- Language-agnostic core with a Python plugin; optional OpenShell sandbox; ships as a Claude Code plugin.