The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted source code during the review process, which is an inherent attack surface for this type of tool. It mitigates risk through a mandatory 'Reviewer-Mode Reset' and 'Review Scoping' process to isolate the diff, and provides detailed security lenses to identify vulnerabilities like secrets exposure and injection attacks.
[COMMAND_EXECUTION]: The instructions mandate the use of git status and git diff to establish review boundaries. These commands are used solely for context gathering and are not combined with unsafe patterns like piping to a shell or using unvalidated user input.
[SAFE]: The skill follows comprehensive security practices, including floors for security-related findings (secrets, access control) and explicit instructions for safely running the reviewer in automated CI environments to prevent privilege escalation from untrusted branches.

code-review