deep-agents-orchestration

What problem does it solve? Building agents that delegate work to subagents, plan multi-step tasks, and pause for human approval requires coordinating several middleware components, and misconfiguring them leads to lost state, failed interrupts, or subagents that silently lack expected tools. ## Core Features & Use Cases - Subagent Delegation: Define custom subagents with their own tools, system prompts, and models, invoked through the built-in task tool, with guidance on statelessness and skill inheritance. - TodoList Planning: Use the write_todos tool to plan and track multi-step work with pending, in_progress, and completed statuses persisted per thread. - Human-in-the-Loop Approval: Configure interrupt_on per tool to approve, reject with feedback, or edit proposed actions before execution, with checkpointer and thread_id requirements. - Use Case: An engineer builds a deployment agent where a code-deployer subagent runs tests and deploys, but the deploy step pauses for human approval before executing against production. ## Quick Start Ask the agent to create a Deep Agent with a custom subagent and human approval required before any file write operation.