The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input from repository files and git metadata, which is interpolated into prompts for the main agent and its spawned subagents. Maliciously crafted PR descriptions or code comments could attempt to influence the agent's behavior during the review or auto-fix process. Ingestion points: git diff (Step 5), gh pr view and git log (Step 3), and TODOS.md (Step 3). Boundary markers: The instructions do not define explicit delimiters or 'ignore' warnings for the interpolated data. Capability inventory: The skill uses Edit and Write to apply auto-fixes (Step 9b) and Bash to execute project commands (Step 11). Sanitization: No sanitization or validation of the untrusted content is performed before it is used in prompts.
[DYNAMIC_EXECUTION]: During the post-fix verification (Step 11), the skill dynamically identifies and executes test runners and linters defined within the project being reviewed (e.g., package.json scripts, Makefile targets, Rakefile). While necessary for the skill's functionality, this involves executing shell commands provided by the repository under review.

review