The Agent Skills Directory

[COMMAND_EXECUTION]: Executes gh CLI commands to manage GitHub repositories, issues, and pull requests. The skill implements a robust risk-classification system (Safe, Write, Destructive, Forbidden) to prevent accidental or unauthorized execution of high-impact commands.
[EXTERNAL_DOWNLOADS]: Facilitates cloning repositories and downloading workflow artifacts from GitHub's official infrastructure. These operations are core to the skill's purpose and target a well-known, trusted service.
[PROMPT_INJECTION]: Ingests potentially untrusted data from GitHub (PR bodies, issue comments, and workflow logs) via commands like gh pr view and gh run view --log. This presents an indirect prompt injection surface.
Ingestion points: gh pr view, gh issue view, gh run view --log in SKILL.md and references/safety-rules.md.
Boundary markers: Absent.
Capability inventory: Full repository management via gh including code modification, secret management, and repository deletion.
Sanitization: Absent.
Risk Assessment: This surface is mitigated by the skill's mandatory safety protocols requiring human confirmation for all non-read operations.

github-cli