Enforces rigorous evidence-based verification before allowing success claims, commits, or pull requests.
This skill establishes a strict 'Iron Law' for Claude, ensuring that no work is claimed as complete without fresh, verifiable evidence. It implements a mandatory gate function that requires identifying proof-of-work commands, executing them, and analyzing the full output before making any assertions of success. By preventing rationalizations and 'hallucinated' completion, it maintains high code integrity and trust between the AI and human developers, particularly during complex tasks involving builds, tests, and linting.
Key Features
01Cross-references agent success reports with actual VCS diffs and logs
021 GitHub stars
03Prevents common rationalizations like 'should work' or 'seems correct'
04Mandates Red-Green cycle verification for bug fixes and regression tests
05Implements a multi-step Gate Function for all status claims
06Enforces the 'Evidence Before Assertions' core principle
Use Cases
01Verifying bug fixes by confirming the failure-to-success transition
02Validating that a build is successful and linter-clean before moving to the next task
03Ensuring all unit and integration tests pass before creating a pull request
