Factory
Pages

6. Explore

Do not plan the product yet. First answer the questions that you cannot answer by talking. This stage is worth the time, because a change now costs a file in the hub. Later, it costs a pull request.

You start with: the Explorations tab, with the questions from page 4.

You finish with: an answer to each question, and a direction that you say is locked.

#What an exploration is

An exploration is one question and one folder. The agent can read the code of the app, and it cannot change it. All that it makes stays in the folder, so a rejected idea does not come back.

Your question is aboutButtonAre you there?
A fact, a tool or a measurement. "Does this sync tool work offline?"BuildNo
How it looks or reads. "How should the phone app look?"ExploreYes, for each round

To add a question, press New, and start with "An exploration:".

An exploration: does PowerSync fit an offline field app on AdonisJS and Postgres? How to find out: build the smallest sync between one Postgres table and a SPA, and measure what happens offline and on reconnect.

#Run a research question

  1. #Press Build

    Open the Explorations tab, click the question, and press Build. You can leave it. The agent makes each decision by itself, and writes it under ## Evidence.

  2. #Read the answer

    When the status is review, open the item. Read the text, then open its files in the Artifacts tab.

  3. #Press Accept

    Press Accept, then Follow in a tab. Choose one of three answers:

    You sayStatusThen
    The findings answer the questiondoneGo to the next question.
    They do not answer itreadyChange the question or how to find out, and run it again.
    Drop itdiscardedThe folder stays, with your reason.

#Run a design question

Press Explore, then Follow in a tab. The session works in this order, and it waits for you at each step.

  1. #References

    Name the products that you want this to feel like, and the ones that you do not. The agent looks at each one, and writes what to take and what to refuse.

  2. #Wireframes

    The agent writes one HTML file for each screen, with no colour. It uses the final words, and shows each state: empty, loading, error and a long list. Open them from the Artifacts tab, and say what is wrong.

    Say "the wireframes are locked" when they are right. Design starts only after this.

  3. #Design rounds

    Each round has two or three concepts that differ in their idea, not only in colour. If you use Claude Design, the agent gives you a prompt, and you bring back the export. Otherwise it makes an artifact or an HTML file.

    Say what is right and what is wrong. Your reasons for a rejected concept go in rejected.md, in your words.

    To compare versions of one part, run /prototype or /variant in the session.

  4. #Choose

    Choose a concept. The agent writes what a build must keep, and sets the status to review. Press Accept.

You can stop at any time. Type /exit, and press Explore again later. It continues from the files.

#When to stop exploring

Stop when you can describe what the first version does, screen by screen, and each answer is written in a finding. You decide this. The factory does not.

Do not explore everything. An unknown that a small task can settle is a task, not an exploration.

#You are done when

  • Each exploration that matters is done or discarded.
  • You can say "the direction is locked".

#Next

7. Plan the work.