verification-before-completion

To stop AI from claiming code is finished without running tests, enforce a verification protocol that requires fresh execution of validation commands and output integrity confirmation before any task completion assertions are made.

Evidence-based task completion is a verification protocol requiring that all success claims are backed by concrete command output, ensuring code changes, test results, or system builds are proven valid through fresh execution before being marked as finished.

To verify bug fixes before reporting them as resolved, run the failing test to confirm it fails, apply the fix, run the test again, and execute validation commands to confirm the output matches expected success criteria.

Yes, you can enforce a verification gate for code changes without external dependencies by mandating a check of commands, exit codes, and output before any success claim can be made, ensuring all status updates are accompanied by command output.

Your AI agent reports success without checking build results because it lacks a strict verification protocol enforcing mandatory fresh execution of validation commands and output integrity confirmation before making any success assertions.