The Agent Skills Directory

[PROMPT_INJECTION]: The skill is designed to ingest and execute instructions from external sources, presenting a surface for indirect prompt injection.
Ingestion points: Instructions are retrieved from the AGENTS.md file in the local repository and from specific sections of GitHub Pull Request comments (labeled "Prompt for AI Agents") authored by CodeRabbit bots (coderabbitai, coderabbit[bot], coderabbitai[bot]).
Boundary markers: The skill lacks explicit boundary markers or delimiters when processing these external prompts; it is instructed to follow the agent prompts literally.
Capability inventory: The agent possesses capabilities to modify local files (via the Edit tool), execute shell commands (git, gh), and perform GitHub API operations such as commenting and reacting.
Sanitization: No validation or sanitization is performed on the content retrieved from AGENTS.md or PR comments before the agent processes them as instructions.
[COMMAND_EXECUTION]: The skill performs several shell-based operations necessary for its workflow:
Executes git commands for status checks, branch identification, commits, and pushes.
Uses the gh (GitHub CLI) to list Pull Requests, fetch thread data via GraphQL, and post summary comments.
Offers to execute build, lint, or test commands as specified in the repository's AGENTS.md file during the validation step.

autofix