What problem does it solve? Orchestrating an autonomous coding agent from another agent or automation pipeline requires handling interactive prompts, permission dialogs, session state, and cost limits, which is error-prone when done ad hoc. ## Core Features & Use Cases - Two Orchestration Modes: Run one-shot non-interactive tasks with claude -p print mode, or drive multi-turn interactive REPL sessions through tmux with explicit dialog handling for workspace trust and permission prompts. - Structured Output & Session Control: Capture JSON results with session IDs, token usage, and cost data, enforce JSON schemas for extraction, and resume or fork sessions across runs. - Use Case: A CI pipeline pipes a git diff into claude -p with --max-turns and --allowedTools to review a pull request for bugs and security issues, returning a structured JSON verdict without any interactive prompts. ## Quick Start Ask the agent to run claude -p with a coding task, an allowedTools whitelist, and a max-turns limit inside your project directory.