The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of the git and gh (GitHub CLI) tools to manage repositories, branches, pull requests, and authentication. It also uses curl to interact with GitHub REST API endpoints.
[EXTERNAL_DOWNLOADS]: Instructions are provided to download GitHub runner packages from the official actions/runner repository. The skill correctly advises verifying the SHA-256 digest of these assets before extraction to ensure integrity.
[PERSISTENCE_MECHANISMS]: The skill includes instructions for maintaining the availability of self-hosted GitHub runners across sessions by installing macOS LaunchAgents via svc.sh and configuring Windows Scheduled Tasks to keep the runner active.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted external data from GitHub, which could contain malicious instructions targeting the agent.
Ingestion points: Data is ingested via gh pr view, gh pr list, gh api calls (e.g., for branches and events), and git diff for code reviews.
Boundary markers: No specific delimiters or "ignore embedded instructions" warnings are documented for the ingestion of these external fields.
Capability inventory: The agent has the ability to execute shell commands (git, gh, curl), write to the filesystem, and perform network operations.
Sanitization: While the skill warns against shell metacharacter injection when writing PR descriptions (using --body-file), it does not specify sanitization or validation logic for the content it reads from GitHub.

github-workflows