The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted code diffs and file content from pull requests. This represents an indirect prompt injection surface where content within the PR could attempt to bypass instructions or influence the agent's behavior during the review process.
Ingestion points: Branch diffs calculated via git merge-base, contents of changed files, and code from callers of changed functions (SKILL.md).
Boundary markers: None explicitly defined to separate untrusted diff content from agent instructions.
Capability inventory: The skill can read local files, modify/write files to fix findings, and interact with the GitHub CLI (gh).
Sanitization: Not explicitly implemented; however, the skill incorporates a human-in-the-loop requirement for sensitive actions.
[COMMAND_EXECUTION]: The skill utilizes standard development tools, including git and the GitHub CLI (gh), to perform its intended functions of branch analysis and pull request management (SKILL.md).
[SAFE]: The skill implements a human-in-the-loop safety mechanism, explicitly requiring a "yes" from the user before executing high-risk operations such as gh pr ready, merging, or enabling auto-merge (SKILL.md).

pr-review