The Agent Skills Directory

[Indirect Prompt Injection] (SAFE): The skill interacts with external content such as pull request descriptions and issue bodies. While this presents a surface for indirect prompt injection, the risk is managed by the use of standard CLI tools for display and processing.
Ingestion points: gh pr view, gh issue list, git diff.
Boundary markers: Absent in instructions.
Capability inventory: Git and GitHub CLI operations via the Bash tool.
Sanitization: Relies on standard CLI tool rendering and GitHub's own safety protocols.
[Data Exposure & Exfiltration] (SAFE): Network activity is restricted to GitHub via the official gh CLI tool, which is a trusted external source. No unauthorized sensitive file access or credential exposure was detected.

git-workflow