The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingest and processes untrusted code diffs as part of its primary review function, creating an inherent but mitigated attack surface for indirect prompt injection.
Ingestion points: External code diffs and user-provided intent descriptions are interpolated into reviewer prompts via the {DIFF} and {INTENT} placeholders in references/reviewer-prompt.md.
Boundary markers: The prompt templates use standard Markdown headers to separate input blocks but do not include explicit instructions for reviewers to disregard embedded commands or instructions within the code content.
Capability inventory: The skill's capabilities are restricted to read-only operations using git and gh tools. It explicitly mandates that the lead synthesizes a verdict without automatically applying code changes.
Sanitization: No specific sanitization or filtering of content within the diffs (such as malicious comments or instruction-laden code) is performed.
[SAFE]: The skill follows security best practices for AI agents by implementing a "report-only" policy and disabling automatic model invocation for sensitive operations. The use of standard development tools (git, gh) to gather metadata is consistent with the stated purpose of code review.

interrogate