Agent Planning And Build Flow

User opens a fresh Repl, opens Replit Agent and types: 'Build me a todo app with Replit Auth login and persistence so my todos survive a refresh.' Agent must produce an editable plan before writing any files.

Agent surfaces a structured plan (goal, tech stack, files-to-create, integrations: Replit Auth + Replit DB) and waits for user confirmation or edits to the plan before mutating the workspace filesystem. Do not start writing code on the first turn — the docs.replit.com/replit-ai/agent flow is plan-f…

User prompt: 'A Streamlit app that reads a CSV from Object Storage and plots a line chart. No login.' Agent's plan adds Replit Auth and a Postgres dependency.

App spec must match the prompt: Streamlit framework, Object Storage as data source, no auth, no extraneous database. Do not silently add integrations the user did not ask for — surface them as suggestions in a separate plan section the user can decline.

After initial plan + first checkpoint produced a working skeleton, user sends a continuation: 'make it use Tailwind for styling.' Agent must extend, not restart.

Treat 'make it' as an incremental request: amend the existing plan, edit only the files needed to add Tailwind (e.g., postcss config, CSS import, Tailwind classes in components), keep prior structure intact. Reuse the same Repl, do not fork.