This is discipline for the moment right before you say "done" or commit your work. It enforces a simple gate: identify the verification command, run it fresh, read the full output, confirm it actually proves your claim, then and only then make the claim. The iron law is no completion claims without fresh verification evidence. It's built from 24 failure memories where Claude shipped undefined functions, missed requirements, or broke trust by claiming success without proof. The tone is blunt because the problem is common: exhaustion, overconfidence, or trusting agent reports leads to saying "should work" instead of running the test. Use this when you catch yourself about to call work complete.
npx skills add https://github.com/davila7/claude-code-templates --skill verification-before-completion