The Agent Skills Directory

[COMMAND_EXECUTION] (MEDIUM): The skill defines a parameter elevated: boolean described as 'Run on host instead of sandbox'. It also explicitly instructs the use of the --yolo flag for Codex, which is described as 'NO sandbox, NO approvals'. These features allow for unconstrained command execution on the host system.
[EXTERNAL_DOWNLOADS] (LOW): The skill documentation includes instructions for installing external packages via npm install -g @mariozechner/pi-coding-agent and routinely uses git clone to pull external repository data for processing.
[PROMPT_INJECTION] (LOW): Category 8 (Indirect Prompt Injection) risk is present because the skill is designed to ingest data from untrusted sources like GitHub PRs and external repositories.
Ingestion points: Git clones and gh pr checkout commands used for 'Batch PR Reviews'.
Boundary markers: None explicitly used to wrap untrusted content, though workdir is used for process isolation.
Capability inventory: Extensive bash capabilities including pty, background execution, and elevated host access.
Sanitization: No sanitization of external code or PR content before it is processed by the coding agents.

coding-agent