Mutation Testing Added as Fourth Merge Gate to Catch Weak Agent-Written Tests

Developers using AI agents to write code face a hidden risk: the same model that writes the code also writes the tests, meaning both can share identical blind spots. Mutation testing addresses this by deliberately injecting small faults into source code and checking whether the test suite detects them, scoring test quality rather than just execution coverage. A lightweight Python harness demonstrates the approach, mutating an implementation file one change at a time and recording whether each mutant is killed or survives the unchanged test suite. In a sample patch, a scoring function's test suite passed all assertions yet still allowed one mutant to survive, revealing a gap in boundary coverage that code coverage metrics would not have flagged. The technique is proposed as a fourth automated merge gate, following property checks, fixtures, and flaky-test freezes, to enforce that agent-generated tests can actually catch regressions.