evidence-labeling-protocol: Claude Skill that forces agents to label the evidence behind a claim

Key points

1. It carries two binding rules: a success claim must be accompanied by the literal output in the same context, and a clean determination such as tests passing or a clear security scan requires a known-positive control plus more than one detection method.
2. The target is the documented failure mode where an agent reports a positive outcome it never checked.
3. Installation notes cover both Claude Code and Claude.ai.