What each agent role produces and how its output is structured

Each of the six specialist roles has a fixed system prompt that shapes the kind of output it returns. Knowing the expected format helps you pick the right role and know what to expect back.

Output by role

• research — a concise markdown brief. Use it to gather facts and context quickly.
• reasoning — a step-by-step, structured analysis that weighs trade-offs. Use it for decisions and comparisons.
• design — hardware/product design notes covering dimensions, materials, components, and cost.
• document — a clean, finished markdown document, suitable as a deliverable.
• patent — structured claims plus a summary, explicitly flagging what needs human/legal review.
• manufacturing — a practical sourcing/BOM-oriented markdown plan.

Everything is markdown

Across all six roles, the output is markdown text generated by the model. That keeps results easy to read inline, save as documents, and feed forward into later steps.

How results reach you

When Camelia delegates, she calls spawn_subagent(role, task), which runs one role-specialized model call and returns its markdown. The result can optionally be saved as a document in a project folder you choose. In a flow, each step similarly runs a role and can save its artifact with a smart label.

Setting expectations

Because each role is driven by its system prompt rather than a rigid template, you'll get consistent style but not field-for-field identical structure every time. For example, a manufacturing result is an AI-authored sourcing/BOM markdown plan, not a catalog export. Pick the role whose described output most closely matches the deliverable you need, and Camelia will route accordingly.