Complete every phase before prompting — so AI output stays identical across every session
AI has no memory between sessions — every new prompt starts from zero. Without complete context, AI guesses everything: libraries, patterns, naming conventions. The result is code that looks different every time even with the same prompt. The fix: inject context before every prompt.
1.1
npx create-next-app@latest my-project \
--typescript \
--tailwind \
--eslint \
--app \
--src-dir \
--import-alias "@/*"1.2
# State & Data
npm install zustand @tanstack/react-query
# Validation
npm install zod
# UI Utilities
npm install clsx tailwind-merge class-variance-authority
# Icons
npm install lucide-react
# Supabase (if using)
npm install @supabase/supabase-js @supabase/ssrCheck every box before prompting AI to do anything new in the project
Copy and fill in the brackets each time
Read `.agent/CONTEXT.md` and `.agent/design-system.md` first
TASK: [create / fix / refactor]
TARGET: [component / hook / function name]
LOCATION: src/components/features/[folder]/
INPUTS: [props or parameters]
BEHAVIOR: [logic description]
CONSTRAINTS:
- Use [specific library] only
- Do not use [unwanted thing]
- Must use named export
OUTPUT: TypeScript + Tailwind CSS