Merge Guard

AI coding agents love to say “Done, all tests pass.” I’ve spent five years building finance integrations, where code that looks done but isn’t moves real money the wrong way. So I don’t trust “done.” I verify it. MergeGuard is a referee between Meta’s Muse Code and your main branch. It re-runs the real tests in a clean worktree and catches the classic cheats: deleted, skipped or edited tests, stubs, and fake passes. Then Muse Spark reads the agent’s own session log and checks every claim it made: verified, false, or no evidence. If something’s wrong, MergeGuard sends it back to Muse Code to fix and checks again. Risky changes, like an auth bypass, go to a human. One rule: real tests decide, not the AI. The AI can add a block, never remove one. Built solo in one day, mostly by Muse Code itself. https://pamantonio.github.io/mergeguard/