AI product Open source

Agent Review

Agent Review is an open-source skill for reviewing agent work, developed by Corey Gallon for Claude Code and Codex. It provides three cumulative review levels: Level 1 requires the agent to identify the expected artifact and verification check, then finish with an ARTIFACT-VERIFIED status and fingerprint or an ARTIFACT-MISSING explanation; Level 2 has two agents independently extract facts, uses a control total, and settles every difference against the original source; Level 3 adds a self-audit, adversarial cold reviewers with explicit reasons to challenge the work, and fresh reviewers who check the revised result. Claims are classified as measured, source-verified, inferred, or assumed, and findings and fixes are independently checked against the source. The skill is installed by cloning its repository into the Claude Code or Codex skills directory and is released under the MIT license.

View repository Visit site Mentioned in 1 video ↓

What Agent Review is used for

2 uses taken from transcripts — each links to the moment in the video.

  • Open-source packaging of the three review levels as agent skills so an agent can check its own work

  • The three review levels — artifact verification, independent extraction with source checks, and adversarial/fresh review — packaged as agent skills and released as open source.

Videos mentioning Agent Review

1 in the library.