Implement in a new session

Continue the last session. When the plan is loaded, tell it to implement the tasks and how you will test.

The idea

You do not start from “what can the model do?” You start from the work already in orkai. You also do not paste overview, the plan, and the Fullstack Developer workflow. Setup already wrote AGENTS.md. That file tells the agent to call overview() first. Overview points at the latest session. The session points at the plan.

After that load, OpenCode tells you what is ready: M1, or M1 and M2, depending on how many milestones and tasks the plan created. By design it works task by task and then stops. It is waiting for you to say it can continue.

Do this in OpenCode

1. Start a new chat in the Stall folder. Do not keep the planning thread.
2. Paste the continue line.
3. You should see orkai calls: overview, the last session, the plan, tasks. Then a confirmation: M1 is ready, or M1 and M2, whatever the plan actually has. It may offer the Fullstack Developer workflow. It is not implementing yet.
4. Paste the implement line. Let it finish every pending task. At the end it should tell you how to test by hand.
5. Do that test yourself. Marking a task done in orkai is not the same as seeing the shop.
6. Save the session. This is daily practice, not a nice-to-have.

Task status fields: Plans, milestones and tasks.

Continue

Let's continue with the previous session.

Implement

OpenCode stops after a task unless you tell it otherwise. This line is normal. It is you confirming it can keep going, and that you want the test steps at the end, not a guess that v1 is done.

Let's implement all the tasks and let me know at the end how to test it manually.

When there is a shop to open

The agent should have told you how to run it. This is what you are checking, not a spec to paste back into the chat.

• Butter croissant: 350 cents, available, allergen gluten and dairy.
• Seasonal soup: 600 cents, sold out, allergen varies.
• Black coffee: 250 cents, available, allergen none.

Run the app the architecture saved (Node.js, SQLite, Next.js unless it recommended something else). Usually npm run dev. You see all three products, including sold-out. Order a croissant: saved, price on the line. Order the soup: rejected. Money is integer cents. No payments, no accounts.

Save the session

When the tasks in this chat are done, save. Always. The next chat continues from orkai, not from this thread. Same habit after planning, after v1, after every slice.

Save the session, I will continue in a new one.

Let's save the session, we are done for today.

Let's save the session to continue the implementation later.

Recap

You continued. orkai held the story. You confirmed the tasks could all run, you tested, and you saved. Next we extend the same requirement, not a new product.