Verification Before Completion

This skill implements a strict 'Iron Law' for Claude, mandating that no status claims, PRs, or commits occur without fresh, documented command output. It prevents the common LLM pitfall of assuming code works based on intent rather than execution, requiring the agent to identify and run specific verification commands—such as tests, builds, or linters—and confirm the exit codes before making any success assertions. By prioritizing evidence over intuition, it builds trust and eliminates the risk of shipping broken code or incomplete features.

主要功能

使用场景