Studio

Studio shows how one agent is wired: its skills, the agents it hands work to, and the exact prompt its model receives. Use it when an agent behaves oddly and no single settings screen explains why.

• Pick an agent in the header selector. The canvas shows its skills and delegates; click a delegate to move to it. Health and Workers are views in the same selector.
• The inspector lists the agent’s model, role, tools and delegates, and the Composed prompt: its own prompt plus each attached skill, in order.
• Edit the prompt or a skill’s text in place. Each save is a new version; History shows a diff and reverts in one click.
• Change the model, settings, skills and delegates. Reset to default puts a built-in agent back to how it shipped.
• The sandbox runs a test conversation against the composed prompt. It saves nothing and runs no tools, so you can try a prompt change safely.

Assistant

The assistant has no tools for Studio. Ask it how Studio works, then make the change here:

• “How is an agent’s composed prompt put together?”
• “What does Reset to default change on an agent?”

Technical

The canvas is drawn from the live database rows, not from the shipped defaults, so it shows your brain as it is now. The Health view runs the config-integrity checks against the system manifest and flags what differs. Studio writes the same agent rows as Settings > Agents. Text versions are kept in prompt_versions, one history per agent prompt or skill, so the history survives model changes and regrants.