The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the official GitHub CLI (gh) to perform repository operations.
Evidence: Includes documentation for gh pr, gh issue, gh run, and gh api commands in SKILL.md.
[PROMPT_INJECTION]: The skill processes untrusted external data from GitHub, creating an indirect prompt injection surface.
Ingestion points: Commands such as gh pr view, gh issue list, and gh run view in SKILL.md ingest data from pull requests, issues, and workflow logs.
Boundary markers: Absent; untrusted content is not clearly delimited from agent instructions.
Capability inventory: The skill can perform actions like gh pr merge, gh pr create, and gh issue create as shown in SKILL.md.
Sanitization: Partial; while some commands use --jq to filter specific fields, the body and title of PRs/issues (which contain free-form text) are still ingested into the agent context.

github