Proof Agent

This skill implements what it claims, but you should be careful before installing or enabling it in sensitive repositories. Key considerations: - Tokens and scopes: The action and scripts expect a GitHub token / Copilot access token. Only provide a token with the minimum necessary scopes and avoid broad account-level Copilot tokens when possible. Use repository-scoped fine‑grained tokens and limit write scopes if you only need comment posting. - Data leakage: verify.sh and the verification prompt include diffs, file lists and commit messages. If secrets or sensitive data are present in those diffs, they will be sent to the external verifier (GitHub Copilot). Do not enable automatic verification on repos with secrets or PII unless you first add scrubbing or gating to remove sensitive content. - Operational test: Try the tool in a non-sensitive test repo first to observe what is sent to Copilot and what tokens are required. Confirm the 'gh' and 'gh copilot' calls behave as you expect in your environment before enabling the action on important repos. - Metadata mismatch: The skill metadata does not declare required credentials but the implementation needs them; treat the absence of declared env vars as a documentation/packaging oversight and assume you will need to supply tokens in the action inputs or environment. - Review and harden: If you plan to use this, consider adding pre-send scrubbing (remove file contents for files matching secret patterns), restrict auto-trigger conditions, and audit the tokens used to ensure they cannot be used to exfiltrate repository data beyond the verifier purpose.