agent-builder

You can draft new agents and attach scheduled automations using a draft-only workflow that prevents immediate publication. This approach enforces a depth-1 cap and requires a draft state for all artifacts, allowing human publishing via the UI later.

Wiring webhooks or app events to trigger agent actions is handled by attaching schedules to agents in a draft state. The workflow provides tools like attach_schedule to compose these triggers while keeping all artifacts unpublished for review.

Yes, you can enforce governance rules like a depth-1 cap and a draft-only lifecycle before human publication via the UI. The workflow ensures all agent artifacts remain in a draft state, preventing unauthorized or immediate deployment.

The primary limitation is a strict depth-1 cap, restricting the complexity of agent hierarchies. Additionally, all artifacts must remain in a draft state, meaning no immediate publication or autonomous execution can occur until manually published via the UI.

You assign specialized skills and MCP servers to drafted agents using the assign_skill and assign_mcp tools. These tools allow you to compose end-to-end draft workflows, extending agent capabilities while preserving the draft state for later UI publishing.

The workflow requires a draft state for all artifacts to enforce controlled governance and prevent immediate publication. This ensures teams can safely compose end-to-end agent workflows with scheduled automations while retaining human oversight before final UI deployment.