The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the Python subprocess module in scripts/run_pi.py and scripts/provider_runtime.py to invoke the pi CLI. Command arguments are constructed using shlex.quote to prevent command injection, and execution is managed with timeouts and controlled input/output streams.
[EXTERNAL_DOWNLOADS]: The skill documentation refers to the pi coding agent CLI, which is an external dependency typically installed via npm. The skill verifies the presence of this tool using shutil.which before attempting execution.
[INDIRECT_PROMPT_INJECTION]: The skill processes external content from files (prompts, images, session context) and interpolates them into model prompts. It includes structural mitigations such as role-based instructions and explicit execution constraints to guide the model's behavior and minimize the impact of potentially malicious data in the processed files.
[DYNAMIC_EXECUTION]: The skill uses sys.path.insert and dynamic imports to access shared utility scripts located in the agent environment's shared/scripts/ directory. This is used for routing, receipt validation, and prompting logic.

pi-runner