PRD Studio

PRD Studio turns one paragraph of an idea into a finished product requirements document. You start it, answer questions, and it hands you a document you can read and export. It is a core Omniscio feature with its own screen.

What it does

You describe the idea in a paragraph. PRD Studio starts one AI session whose only job is to interview you, and it works through twelve planning steps in a fixed order:

1. Product Vision
2. Feature List
3. User Journey
4. Technical Hurdles
5. Design Language
6. Tech Stack
7. Data Flow
8. User Red Team
9. Developer Red Team
10. Designer Red Team
11. Manual Work
12. Final Improvement

Steps 8 to 10 are three critiques of the earlier work, and they are optional: you choose at creation whether a PRD includes them, and that choice is locked afterwards. A PRD without them simply skips those steps.

Each step ends as one stored section. When every step that was not skipped holds a section, the PRD is finished and can be exported as one markdown document.

How the interview works

The AI asks one question at a time, with two to five options and usually a recommendation. The question appears as a card on the PRD's screen; you can also answer it by typing in chat. Either way, the answer is stored the moment it arrives and is handed to the AI once.

Two things you will notice:

• The AI never decides what comes next. The app owns the step order. The AI asks the app what to do before every move, so it cannot skip a step and cannot finish a PRD that is missing a section.
• Answers can wait. If the AI's session has stopped, your answer is still stored. The PRD shows "Waiting on the AI" and offers Resume, which starts a fresh session at the step you were on with every earlier answer and section already loaded.

Size limits

Every stored piece has a ceiling, so a long document can never overflow what the AI can accept mid interview: a question is capped at 1,000 characters, an option label at 200 and its detail at 600, a reason at 1,000, your answer at 4,000, a step summary at 1,500, and a section at 16,000. Each step also allows a fixed number of questions (18 for the Vision step, fewer for the rest); once the budget is spent the AI must write the section rather than keep asking.

Your controls

• New PRD — give it a title and your idea (a sentence is enough; at least 20 characters) and choose whether to include the red-team steps. One session starts immediately and asks the first question.
• Answer — on the card or in chat.
• Pause — stops the AI session and keeps everything stored. Resume starts a fresh session at the same step.
• Cancel — stops the PRD and its session for good. Nothing more can be written to it.
• Delete — removes the PRD and everything it stored, behind a confirmation.
• Export — writes the whole document as a markdown file named after the PRD title.
• The screen shows the session's cost using the same session cost the rest of the app shows.

What it is not

PRD Studio does not split the document into build tasks, build the product, or deploy anything. Foundry, the older PRD plugin, keeps working unchanged and is a separate thing.

For developers

The contract is prd-studio-contract.md and the map is prd-studio-map.md — tables, the twelve-step state machine, the /next hand-off, the five agent routes and the IPC surface the screen uses.