The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the GitHub CLI (gh) to interact with pull request data and executes a local Node.js script (extract-pr-comments.js) to parse and organize feedback into actionable tasks.
[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface by ingesting untrusted data from external sources (PR comments) and presenting it to the agent as direct instructions. Ingestion points: GitHub PR review and issue comments are fetched via gh and processed into a TODO list. Boundary markers: No specific delimiters or "ignore previous instructions" warnings are applied to the extracted comment content. Capability inventory: The agent has permissions to modify files, execute build/test commands, and perform Git operations. Sanitization: The extract-pr-comments.js script performs classification based on keywords but does not sanitize or escape the content of the comment bodies before they are presented to the agent.

pr-review