The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on the GitHub CLI (gh) to perform repository management tasks. These commands include gh pr create, gh pr edit, gh pr comment, and gh pr view. These are legitimate operations for a PR management skill and are used with appropriate flags like --body-file to ensure safe handling of multi-line text.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and act upon configuration files and templates found within the user's repository, which represents a surface for indirect prompt injection.
Ingestion points: The skill reads PR templates, .coderabbit.yaml, .coderabbit.yml, and greptile.json from the repository to determine review policies and PR structure.
Boundary markers: The instructions do not explicitly define delimiters for these external inputs, though they do instruct the agent to remove irrelevant prompts from templates.
Capability inventory: The agent has the capability to write to the repository's PRs and comments via the gh CLI.
Sanitization: The skill advises the agent to verify the diff and only include sections with real content, providing a basic layer of manual-style sanitization by the LLM.

create-pr