Writes the context file an AI coding agent reads before it touches your code.
Answer eight questions about your project and get a finished AGENTS.md and
CLAUDE.md to paste into your repo. Nothing is stored, nothing is sent
anywhere, and there is no model call. The whole thing is assembled in your
browser.
An agent that has not been told what your project is will infer it from file names and get it wrong. One that has not been told what looks deliberate will helpfully undo it. The eight questions are not a survey, they are the eight things that have actually caused a mistake often enough to be worth writing down.
The most valuable one is the third: what looks wrong but is deliberate. Almost nobody includes it, and every codebase has decisions that read as bugs.
The eighth exists because commands and obligations are different things.
build_runner and prisma generate are not things you run when you feel like
it, they are things that must happen after a particular change or nothing
compiles. An agent reads a bare command as optional. It reads a trigger as a
rule.
One markdown file, about twenty minutes, no cloning required. See CONTRIBUTING.md.
Currently supported: Next.js, Python FastAPI, Flutter.
npm install
npm run devThat is the only command. No database, no API key, no accounts, nothing to configure.
npm test # assembly tests, on node:test
npm run build # static export to out/
npm run lint- No backend.
output: 'export'is set so that a server cannot be added quietly later. An API route with request-time code fails the build. - No dependencies beyond the framework.
next,react,react-dom, and that is the list. Tests run on Node's built-in test runner. - Templates are markdown on disk.
templates/index.tslists them. Delete one and the build fails naming the file, so a broken registry cannot ship.
The plan this was built against is in PROJECT.md, and the build notes, including everything that went wrong, are in NOTES.md.