Show HN: Pairmark, race Claude Code vs. Codex on your repo, blind cross-judged (github.com)

A new tool called Pairmark allows developers to run AI coding agents Claude Code and Codex simultaneously on a coding task within their own repository, generating a detailed report that assesses which agent performed better based on objective criteria. By executing a simple command, developers can observe both agents working in real-time within isolated worktrees, with each task receiving the same brief and subject to identical checks. The results include a verdict that highlights the winning agent and the rationale behind the decision, as well as comprehensive performance metrics. This development is significant for the AI and machine learning community as it provides a concrete way to compare the capabilities of different coding agents using real codebases rather than subjective opinions. It promotes empirical evaluation by offering detailed reports that include metrics such as execution time, code changes, and test results, thereby helping developers make informed decisions about which AI tool best suits their coding needs. Pairmark operates entirely on the local machine without external dependencies, ensuring privacy and security by keeping sensitive data contained.