The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted external data from GitHub PR comments generated by third-party bots (Codex, CodeRabbit, etc.), which could serve as a vector for indirect prompt injection attacks.
Ingestion points: Fetches reviewer findings, inline comments, and issue comments using gh api calls in the workflow defined in SKILL.md.
Capability inventory: Possesses the capability to read project files and use the Edit and git commit tools to modify the codebase based on the external findings.
Boundary markers: The skill relies on regex-based normalization but does not explicitly detail the use of secure delimiters or specific instructions for agents to ignore embedded commands in the bot output.
Sanitization: Implements a robust human review step (AskUserQuestion) requiring explicit user confirmation for each proposed fix before it is applied to the repository.
[COMMAND_EXECUTION]: The workflow relies on the execution of shell commands using the GitHub CLI (gh) and Git to manage PR interactions and code commits.
Evidence: The skill uses gh pr view, gh api, git remote get-url, and git commit within its workflow steps in SKILL.md.

kw-pr-review