The Agent Skills Directory

Command Execution: The skill instructs the agent to execute shell commands such as git diff, gh pr checkout, and npm run preflight. While these are standard tools for developers, executing project-specific scripts like npm run preflight means the agent runs code defined within the repository being reviewed. This is a common pattern for verification workflows but relies on the integrity of the project's own scripts.
Indirect Prompt Injection Surface: The skill processes untrusted data from external sources, such as git diffs and Pull Request descriptions, which could contain instructions intended to influence the agent's behavior.
Ingestion points: Remote PR descriptions, comments, and file changes accessed via gh pr checkout and git diff (referenced in SKILL.md).
Boundary markers: The instructions do not currently include explicit delimiters or specific instructions for the agent to treat the code and PR metadata as data rather than instructions.
Capability inventory: The agent has the capability to execute shell commands (git, gh, npm) and read file contents within the project environment (referenced in SKILL.md).
Sanitization: The workflow does not specify sanitization or filtering of the content retrieved from Pull Requests before analysis.

code-reviewer