After this step: you can take one small stage from assignment to checked behavior.
Choose one piece that can be used end to end. For the ideas app it can be: I type an idea, save it, and see it after reopening. Agree with the agent how you will know that it works. Then watch a demo and report corrections. When you accept the behavior, the agent tidies the code, runs checks, and records the result in the plan.
Prompt for your agent
Complete the next agreed stage from the plan: [stage]. Read requirements, instructions, interface rules, and accepted graphics from the plan. Use shared elements; do not invent each screen’s appearance again. If the project has no screens, skip appearance checking. Confirm the expected result and way to check it, then complete this scope. Show me the working piece in a local preview and compare it with the agreed scenario and screen design where relevant. Fix feedback as it comes; check risky changes to logic, data, and permissions immediately. After I accept the whole demo, tidy the code, do the needed refactor and required tests, and prepare an independent review. Record the result and remaining matters in the existing plan. Do not publish or send changes without my instruction.
Record: the plan contains status, feedback, version, and evidence; checked run instructions go to README or development.md. Check: you can perform the scenario yourself, and the agent separates the result from checks it did not perform.