The Agent Skills Directory

[Defensive Security Guidelines]: The instructions include a dedicated security section that directs the agent to evaluate trust boundaries, path traversal risks, and potential secret disclosure. This proactive approach significantly reduces the risk of overlooking security regressions during the review process.
[Untrusted Code Execution Restriction]: The skill explicitly forbids the agent from executing code within a Pull Request to verify behavior, directing it instead to rely on available artifacts and demonstrations. This is a key safety control to prevent remote code execution in the agent's environment.
[Indirect Prompt Injection Awareness]: By processing Pull Request content, the agent is exposed to untrusted input. The skill addresses this by instructing the agent to 'challenge the change's assumptions' and verify claims against authoritative evidence, which serves as a conceptual mitigation against deceptive or malicious content embedded in PRs.
[Standard Resource Access]: The skill references standard project files such as API baselines and diagnostic descriptors. These are intended for context-gathering within the scope of a repository maintainer's role and do not involve unauthorized data access.

code-review