The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses shell commands like gh (GitHub CLI) and git to manage pull request metadata, fetch comments, and apply code changes. These commands are executed within the local repository context and are necessary for the skill's primary functionality.
[PROMPT_INJECTION]: An indirect prompt injection surface is present as the skill ingests feedback from external GitHub Pull Request comments which could contain adversarial instructions.
Ingestion points: Data is fetched via gh api calls from inline code comments, review summaries, and general PR conversation threads.
Boundary markers: The skill architecture incorporates a mandatory 'Wait for the user to confirm' step before applying any code fixes or posting automated replies, providing a critical human-in-the-loop security control.
Capability inventory: The agent has the ability to modify local files, perform git commit and git push operations, and post comments back to GitHub via gh api.
Sanitization: A triage phase is included where the AI agent is instructed to evaluate the validity of each comment against the current source code before presenting recommendations to the user, acting as a preliminary verification layer.

address-review