Planning the App
This is the session that decides how the whole app gets built. Working in plan mode, the agent pulls in the project rules, the skills and the design system, asks a long run of clarifying questions, and produces an implementation plan. Then, before approving anything, we send sub-agents in to pull that plan apart and improve it. The result gets saved into the project so you can pick the build back up whenever you want.
What you'll walk away with:
- Why plan mode matters, and what you actually have in hand by the end of the session
- Invoking the start-an-app skill and describing the app in plain language
- How CLAUDE.md, AGENTS.md and design.md feed context into the planning session
- What lives in the skills folder, covering auth, databases, AI capabilities and payments
- Swapping out parts of the tech stack, such as replacing Better Auth with Clerk
- Answering the agent's clarifying questions: redirect behaviour, supported QR content types, per-code actions and styling controls
- Choosing Postgres over SQLite, and when SQLite is the right call instead
- Picking the auth method, the download formats, and whether to apply usage limits
- Never trusting the first plan: kicking off three deep-dive sub-agents to review it
- What those review agents look at, including rendering architecture, design system and UX, and security and scope
- Inspecting what each sub-agent is doing, and why the harness you use doesn't matter
- Saving the plan to a plans folder at project level rather than user level
- Resuming an interrupted build by pulling the saved plan back into the chat
- Clearing the conversation afterwards to free up the context window