소개
The Verification Guardrail skill implements a non-negotiable 'evidence before claims' policy for Claude, preventing premature declarations of success. It mandates a 5-step validation process—Identify, Run, Read, Verify, and Claim—across all development tasks including testing, building, and linting. By eliminating assumptions and speculative language like 'should work' or 'seems to,' this skill ensures that every reported fix or feature is backed by fresh command output and verified exit codes, significantly increasing reliability and trust in AI-driven development workflows.