The Agent Skills Directory

[SAFE]: The skill implements a robust security auditing process with clear exclusion criteria and multi-pass verification (filtering and exploit-scenario passes) to ensure high signal quality and minimize false positives.
[COMMAND_EXECUTION]: Executes standard git commands including git status, git diff, and git log to gather context about branch changes. These are read-only operations restricted to the local development environment and are appropriate for the skill's intended purpose of reviewing code changes.
[PROMPT_INJECTION]: The skill processes untrusted data in the form of git diffs (ingestion point: git diff output in SKILL.md). While it lacks explicit boundary markers or automated sanitization, the risk from indirect prompt injection is minimal as the capability inventory is limited to read-only git operations and providing analysis to the user, with no network access, file-write permissions, or arbitrary code execution capabilities.

security-review