Verify

This is the skill that stops Claude from saying "I've fixed the bug" when it hasn't actually run the tests. It enforces a five-step gate before any completion claim: identify the proof command, run it fresh in this session, read the full output including exit codes, confirm it proves the claim, then report with evidence. No "should work" or "probably fixed" allowed. It includes a regression verification protocol for bug fixes that requires the test to fail without the fix and pass with it, proving the test actually catches the issue. The forbidden words list alone (should, probably, seems to, likely) is worth internalizing. If you've ever merged code based on an AI's confident assertion only to find it broken, this enforces the discipline you wished you'd had.

Install to Claude Code

npx -y skills add codeaholicguy/ai-devkit --skill verify --agent claude-code