This is for when you're about to say "tests pass" or "build works" without actually running the commands. It forces a gate function: identify the verification command, run it fresh, read the full output, then make your claim with evidence. Covers the common traps like trusting agent reports, accepting partial checks, or thinking "should work" counts as verification. The tone is blunt because it's addressing a pattern where fatigue and optimism lead to claiming completion without proof. Most useful right before commits, PRs, or moving to the next task when you're tired and want to be done.
npx skills add https://github.com/secondsky/claude-skills --skill verification-before-completion