The replay test

The replay test is simple: could someone recreate your reasoning from just the commit message? Not the code — the why, the approach, and the context behind the change.

The research

We analyzed 2,271 commits from AI coding agents across real-world repositories. The results:

Claude Code 28% pass
Cursor 17% pass
Copilot Agent 14% pass

A "strong pass" means the commit message included why, approach, and context — the three elements needed to understand a change without reading the diff.

The problem

None of the agents write structured why by default. They describe what changed — which you can already see in the diff — but not why it changed, what was considered, or what tradeoffs were made.

Muselet fixes this by giving agents a clear format and validating it at commit time.