Lovable

Build me a local brief-to-app generator to replace Lovable. Requirements:

- A Node 22 CLI: `vibe new "a habit tracker with a weekly grid"` scaffolds a Vite +
  React + TypeScript app into ./apps/<slug>, then edits it in a loop. Generated apps
  persist to plain JSON or SQLite via better-sqlite3.
- Call the Anthropic or OpenAI API with the key from .env and ask for whole files
  back as JSON, path plus contents, never prose I have to dig out of a code fence.
- Every round is a git commit on a branch and prints the diff first. `vibe undo` is
  a git revert, so nothing the model does is unrecoverable.
- After each round run `npm run build` and `tsc --noEmit`, feed failures back, retry
  three times, then stop and show me the error instead of looping.
- `vibe preview` runs the Vite dev server on localhost:5173. No accounts, no
  telemetry, nothing leaves my machine except the model calls.
- Keep the transcript, file tree, and last diff in .vibe/session.json inside the app,
  resend only the files I named, and print tokens and cost per round.
- Out of scope: hosted previews, one-click deploy, and team collaboration. Do not
  build a web UI, the terminal is the interface.
- README: which key to set, what a typical app costs, and the honest warning that the
  output is only as good as the model behind it.

# AGENTS.md — Build guide for Lovable

## Project scope
Send a plain-language brief to a model API, have it write a small React app into a local folder, run it, and keep iterating in the same conversation.

Catalogue verdict: no. The wrapper is easy: a CLI that sends your brief to a model, writes the files, and starts the dev server is one sitting of work. The model doing the writing is the actual product, and you cannot rebuild that. Everything Lovable puts around it, the hosted preview, the deploy, the database wired up for you, and a non-developer being able to use any of it, goes with the subscription.
Use the implementation prompt below to define the deliverable. Complete each phase's acceptance checks before extending the scope.

## Working agreement
- Inspect the repository and its existing instructions before choosing paths, dependencies or commands. Keep one coherent stack and explain changes to the proposed architecture.
- Plan a vertical slice that accepts a real input and produces the useful output described below. Persist only the state the prompt calls for; respect memory-only and upstream-managed workflows. Use fixtures only when they are clearly labelled.
- After scaffolding, document the actual install, development, check and build commands in README and keep them synchronized with the package or project manifest. Do not report commands as successful unless they ran.
- Work in small steps. At handoff, list implemented flows, checks actually performed, remaining blockers, and any credentials or provider setup the owner must supply.
- Do not publish, spend money, contact customers, delete source data or run irreversible migrations without the project owner's authorization.

## Prerequisites
- Node 22
- Anthropic or OpenAI API key
- git

## Stack and architecture
- A Node 22 CLI: `vibe new "a habit tracker with a weekly grid"` scaffolds a Vite + React + TypeScript app into ./apps/(slug), then edits it in a loop. Generated apps persist to plain JSON or SQLite via better-sqlite3.
- A Node 22 CLI: `vibe new "a habit tracker with a weekly grid"` scaffolds a Vite + React + TypeScript app into ./apps/(slug), then edits it in a loop. Generated apps persist to plain JSON or SQLite via better-sqlite3.
- Interface for this AI coding agents and developer workspaces workflow: a CLI with a dry-run or preview command and an explicit output path.

## Security and data integrity
- Every round is a git commit on a branch and prints the diff first. `vibe undo` is a git revert, so nothing the model does is unrecoverable.
- Keep API keys and OAuth secrets in .env; redact them from logs, generated files, and error responses.
- State the practical limit: the frontier model, which is the actual product.

## Agent implementation rules
- Project rule, data: A Node 22 CLI: `vibe new "a habit tracker with a weekly grid"` scaffolds a Vite + React + TypeScript app into ./apps/(slug), then edits it in a loop. Generated apps persist to plain JSON or SQLite via better-sqlite3.
- Project rule, behavior: Keep the transcript, file tree, and last diff in .vibe/session.json inside the app, resend only the files I named, and print tokens and cost per round.
- Project rule, recovery: Every round is a git commit on a branch and prints the diff first. `vibe undo` is a git revert, so nothing the model does is unrecoverable.

## Optional agent skills and references
No installable skill is listed for this guide; use the project rules and the stack documentation.

Read the linked SKILL.md and its dependencies before adding a skill. Select only the skills matching this project's runtime and task; their documentation does not supply API access, credentials or approval to perform external actions. Pin the reviewed revision where the tool supports it. Follow the chosen agent's documented project-level installation mechanism.

## Distribution ideas
These are optional planning notes. Obtain the owner's approval before publishing or contacting anyone.
- Demonstrate this AI coding agents and developer workspaces operation with sample or user-owned data: Keep the transcript, file tree, and last diff in .vibe/session.json inside the app, resend only the files I named, and print tokens and cost per round.
- Show a concrete result or export: Prompt a full web app into existence, with hosted previews and one-click deploy.
- State the practical limit: the frontier model, which is the actual product.

## Engineering roadmap
1. Phase 1, architecture and data: A Node 22 CLI: `vibe new "a habit tracker with a weekly grid"` scaffolds a Vite + React + TypeScript app into ./apps/(slug), then edits it in a loop. Generated apps persist to plain JSON or SQLite via better-sqlite3. A Node 22 CLI: `vibe new "a habit tracker with a weekly grid"` scaffolds a Vite + React + TypeScript app into ./apps/(slug), then edits it in a loop. Generated apps persist to plain JSON or SQLite via better-sqlite3.
2. Phase 2, implement: Keep the transcript, file tree, and last diff in .vibe/session.json inside the app, resend only the files I named, and print tokens and cost per round.
3. Phase 3, implement: Send a plain-language brief to a model API, have it write a small React app into a local folder, run it, and keep iterating in the same conversation.
4. Phase 4, review and output: Prompt a full web app into existence, with hosted previews and one-click deploy.
5. Phase 5, recovery and acceptance: Every round is a git commit on a branch and prints the diff first. `vibe undo` is a git revert, so nothing the model does is unrecoverable. Verify this invariant with a saved fixture: A failed tool call or interrupted run must retain the branch and trace; no file outside the selected repository may change. State the practical limit: the frontier model, which is the actual product.

## Paid-product capabilities outside this build
- the frontier model, which is the actual product
- hosted preview URLs and one-click deploy
- a database and auth wired up for you
- a non-developer being able to ship anything at all

## Implementation prompt
Build me a local brief-to-app generator to replace Lovable. Requirements:

- A Node 22 CLI: `vibe new "a habit tracker with a weekly grid"` scaffolds a Vite +
  React + TypeScript app into ./apps/<slug>, then edits it in a loop. Generated apps
  persist to plain JSON or SQLite via better-sqlite3.
- Call the Anthropic or OpenAI API with the key from .env and ask for whole files
  back as JSON, path plus contents, never prose I have to dig out of a code fence.
- Every round is a git commit on a branch and prints the diff first. `vibe undo` is
  a git revert, so nothing the model does is unrecoverable.
- After each round run `npm run build` and `tsc --noEmit`, feed failures back, retry
  three times, then stop and show me the error instead of looping.
- `vibe preview` runs the Vite dev server on localhost:5173. No accounts, no
  telemetry, nothing leaves my machine except the model calls.
- Keep the transcript, file tree, and last diff in .vibe/session.json inside the app,
  resend only the files I named, and print tokens and cost per round.
- Out of scope: hosted previews, one-click deploy, and team collaboration. Do not
  build a web UI, the terminal is the interface.
- README: which key to set, what a typical app costs, and the honest warning that the
  output is only as good as the model behind it.

## Completion evidence
Demonstrate the prompt's acceptance scenarios against the scoped workflow. Include setup from a clean checkout and failure recovery. Check persistence across restart and export/restore only for the state the prompt says to store; for memory-only tools, confirm that temporary content is discarded as specified. Record actual results and remaining limitations. A detailed plan alone does not establish a working replacement.