The Agent Skills Directory

[COMMAND_EXECUTION]: The skill performs multiple shell operations using git and gh. It interpolates variables derived from user-provided arguments and command outputs into shell strings for branch management, commits, and PR creation.
[PROMPT_INJECTION]: The skill identifies an indirect prompt injection surface when reading existing PR titles and bodies from GitHub. It includes explicit instructions for the agent to treat this third-party metadata as untrusted and to ignore any commands or instructions found within it.
[EXTERNAL_DOWNLOADS]: The skill interacts with GitHub (github.com), a well-known and trusted service, to push code and manage pull requests.
[COMMAND_EXECUTION]: The documentation indicates that the skill may require lifting sandbox restrictions (e.g., dangerouslyDisableSandbox: true) to allow the GitHub CLI to access the OS keyring for authentication, which is a significant permission request necessary for its intended function.
[PROMPT_INJECTION]: The skill employs secure scripting practices by using quoted heredocs (cat <<'EOF') when writing PR bodies, preventing the local shell from interpolating or executing content extracted from git logs or user input.

ship-it