The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from GitHub Pull Requests, such as titles, descriptions, and comments, which represents a potential attack surface.
Ingestion points: Untrusted content is ingested from GitHub via gh pr view and gh api commands in SKILL.md.
Boundary markers: The skill does not use specific structural delimiters for the ingested data, but it enforces a mandatory human-in-the-loop check.
Capability inventory: The skill possesses write capabilities through gh pr review and gh pr comment commands.
Sanitization: The 'Approval Safeguard' serves as the primary mitigation, ensuring that a human reviews all generated draft comments before they are posted to GitHub.
[COMMAND_EXECUTION]: The skill utilizes the GitHub CLI (gh) to interact with repositories.
Scope: The commands listed (view, diff, checks, review) are standard tools for the PR review process and are used according to their intended design.
Access Control: The instructions strictly forbid the agent from executing write commands without explicit prior user approval, minimizing the risk of unauthorized actions.

reviewing-prs