The Agent Skills Directory

[COMMAND_EXECUTION]: The skill documents the use of gh codespace ssh for remote terminal access and gh alias set for creating custom commands that can execute shell scripts using the ! prefix.
[DATA_EXFILTRATION]: Instructions include commands like gh secret set, gh variable set, and gh ssh-key add which are used to transmit sensitive credentials or configuration data to GitHub.
[CREDENTIALS_UNSAFE]: Documentation covers gh auth status --show-token and gh auth token which can output sensitive authentication tokens to the terminal context, as well as commands that access local sensitive files like ~/.ssh/id_ed25519.pub.
[PROMPT_INJECTION]: As the skill processes untrusted external data from GitHub (such as issue comments, PR descriptions, and workflow logs) and possesses significant administrative capabilities (repository deletion, secret modification), it presents a surface for indirect prompt injection. Ingestion points: gh pr list, gh issue view, gh run view --log. Boundary markers: None explicitly enforced in documentation. Capability inventory: gh repo delete, gh secret set, gh codespace ssh, gh api. Sanitization: Relies on standard CLI tool behavior.

gh-cli