Multi-Agent Orchestration
Watch specialized design, development, and quality agents collaborate.
Overview & Objectives
Multi-agent workflows divide a larger task into role-based work and review. Clear scopes, inspectable changes, and a human review of the final result matter more than the number of agents.
Core Tutorial Steps
Understand Specialized Agent Roles & Scopes
Define each role by its deliverable and boundaries. A design role can outline components, a development role can implement a change, and a review role can inspect the proposed result.
Inspect the activity log
Use the workspace activity log to review visible task updates and tool invocations. The log is not a complete explanation of every model operation.
Set a review checklist
Tell the review role which requirements to check, such as changed files, formatting, error handling, or a specific test command. Review results yourself.
Keep iteration bounded
If a check fails, give the relevant error to the implementation role and ask for a focused correction. Re-run the check and stop when the result meets the task criteria.
Customize Workspace Directives & Goals
Inject custom system prompts or behavioral guidelines to steer the entire agent network.
Treat agent output as proposed work: inspect the changed files and run the checks that matter before relying on the result.