Act as the lead engineer on my {{tech_stack}} project. Write the rules file my AI tool reads before every task: Lovable's project knowledge, a Cursor rules file, CLAUDE.md, or AGENTS.md for Codex and others. Read the codebase first, write only what's true of this project, and tell me where you're guessing. Cover, in short imperative lines: - Stack and versions: framework, router, styling, component library, database, auth, hosting. Name the packages already used for forms, dates, data fetching and icons, so the AI never adds a second one - Commands: install, dev, build, lint, typecheck and test, and which must pass before a task counts as done - Structure: where pages, components, hooks, server code and migrations live, and how files are named - Conventions: how data is fetched, how errors and loading and empty states are shown, how forms validate - Design: tokens to use and never hard-code (colors, type scale, spacing, radius, shadows) for this look: {{visual_style}}, set in {{brand_fonts}}. Reuse existing components before making new ones - Ask before touching: auth, payments, access rules and policies, applied migrations, environment files, generated files, and these areas: {{page_or_feature}} - Security: secrets stay on the server, input is validated on the server, and access rules or webhook checks are never weakened to make an error go away - Working style: explain the plan first, keep diffs small, run the tests after every change, stop and ask after two failed attempts Keep it under 150 lines; long rules files get ignored. Give me the file, a 5-line version for tools with tight limits, and where to save each one.