Skip to content

noahvincze/truegreen

v0.1.0MIT

Keeps green checks honest: when a check fails, the agent fixes the code or reports the problem, instead of quietly editing tests, inventing data or swallowing errors.

GitHubOpen Website

Changelog

Unreleased

  • Codex plugin manifest (.codex-plugin/plugin.json) and a portable Agent Plugins manifest (plugin.json); README install steps for Codex and, via npx skills, Cursor, Kimi Code, OpenClaw and others.

  • The audit script as a GitHub Action (action.yml), a pre-commit hook (.pre-commit-hooks.yaml) and a truegreen-audit command (pyproject.toml).

  • Tests that the version and name match across all manifests.

  • README: before/after image, install near the top, Contributing section.

  • docs/why.md: the case for the skill, with sources; README intro rewritten around who decides when a test changes.

  • audit_diff.py skips its own file, so a vendored copy does not flag its own patterns.

  • Tests for the audit script: 19 scenario files in tests/audit_cases/ plus CLI tests.

  • Reproducible evals: 6 cases with hidden tests, reference solutions and cheat overlays, and evals/run.py to run and grade agents with and without the skill.

  • GitHub Actions CI: tests and eval checks on Linux, Windows and macOS, plugin validation.

0.1.0 - 2026-11-01

  • First packaged release: truegreen skill with scripts/audit_diff.py.
  • Repo laid out as a Claude Code plugin and marketplace (.claude-plugin/, skills/<name>/).
  • MIT license.