The Agent Skills Directory

[SAFE]: The skill implements a robust process supervision framework with multiple security layers. It uses private state directories with restricted permissions and protects inter-process communication via Unix-domain sockets using random nonces for authentication. The supervisor performs strict provenance and identity checks on manifest files and process snapshots, verifying PIDs, process group IDs, and start times to prevent session hijacking or manipulation of run state.
[COMMAND_EXECUTION]: The script manages the lifecycle of external processes using standard system utilities. It spawns the pi executable and tools like ps to monitor agent health and execution status. The worker subagent role is configured with bash and edit capabilities, which are required for its documented primary purpose of implementing code changes within an authorized directory.

use-pi-subagents