The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses subprocess.run to call git, tmux, and claude binaries to automate the creation of git worktrees and the management of persistent interactive sessions.
[PROMPT_INJECTION]: The spawn.py script is configured to launch sessions with the --dangerously-skip-permissions flag, which bypasses standard interactive user confirmation for tool execution within those sessions.
[PROMPT_INJECTION]: The monitoring functionality in monitor.py presents an indirect prompt injection surface where content from a child session can manipulate the orchestrator.
Ingestion points: Terminal output captured from tmux panes in scripts/monitor.py.
Boundary markers: None; the script parses raw terminal output using regex patterns.
Capability inventory: The skill provides full control to send arbitrary keyboard input to child sessions and read their entire screen content.
Sanitization: The detection of interactive states (e.g., AskUserQuestion) relies on regex matching that can be triggered or spoofed by content printed to the terminal within the child session.

kage-bunshin