What the patent agent flags for human and legal review

The patent specialist is one of the six sub-agent roles, and its job is to produce structured, IP-flavored analysis — while being explicit about where human judgment is still required.

What the patent agent produces

The patent role's fixed instruction is to return structured claims plus a summary, and to flag what needs human and legal review. So its output is shaped like the bones of a patent-style analysis: claims laid out in structure, a summary of the invention, and clear callouts where a person — ideally a qualified attorney — must weigh in.

Why it flags rather than decides

Patent work has real legal consequences, and an AI model is not a substitute for counsel. The role is built to surface the things that require human and legal review rather than to present its output as a finished, filed-ready, or legally vetted document. Treat its claims and summaries as a structured starting point for a professional to evaluate.

How it fits the workflow

Like the other roles, the patent agent runs via spawn_subagent and relies on the model's knowledge rather than live web search. Its markdown result can be saved into a project folder — in flows, patent-style output is given a "Legal" smart label — so it lands alongside your other documents and can be exported in a bundle.

Using it well

Use the patent agent to organize prior-art thinking and draft claim structure early, then route anything it flags to qualified legal review before relying on it. The agent is designed to help you prepare, not to replace the human and legal review it explicitly calls out.