The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local CLI tools including pi and gh (GitHub CLI) via shell commands to process code reviews and resolve pull request metadata.
[INDIRECT_PROMPT_INJECTION]: The skill accepts user-provided inputs for targets (PR numbers or ref ranges) and model names, then interpolates these values directly into both a shell command and a natural language prompt for a sub-session.
Ingestion points: User-supplied arguments for PR number, ref range, and model defined in SKILL.md.
Boundary markers: None. The input is passed verbatim into the pi command string and its nested prompt.
Capability inventory: Subprocess execution of pi, gh, and git commands.
Sanitization: The instructions explicitly state to pass the user's model 'verbatim' and perform 'substitution' for the target without defining escaping or validation logic, which could lead to command injection or prompt manipulation in the sub-session.

second-opinion