← back to the overview
Beyond generate

The commands you run after the first one

Generating the repo is one command. Keeping it current — adding a technology, implementing a feature, reviewing what an assistant wrote — is the rest of them.

The problem

An empty repo makes a good assistant guess

It doesn't know your auth provider, your folder conventions, or that you never scaffold a router. So it invents them — differently every session. Scaffolding isn't optional context; it's the whole input.

my-app/ — before4 files
src/
package.json
tsconfig.json
README.md
nothing for the assistant to read
my-app/ — after one run+40 files
.cursor/rules/ 9 rules
.claude/skills/ 9 skills
docs/ setup · architecture · roadmap · costs …
prompts/ reusable, stack-aware
checklists/ release, plus per-module
.env.example every var, documented
AGENTS.md GEMINI.md CLAUDE.md
ai-project.config.json replayable
src/ package.json tsconfig.json README.md
Beyond generate

Day one is the easy part

Nobody picks their whole stack up front. Six commands keep a generated project current without ever overwriting something you edited by hand.

implement <feature>the interesting one

Writes a plan, prompts, a checklist and skeleton files for one feature — tailored to the provider your project selected. Same command, genuinely different output.

A project on Supabase Auth
plan.md
→ session persistence via AsyncStorage
→ the RLS policies that gate access
src/features/auth/authClient.ts
src/hooks/auth/useAuth.ts
The same command, on Clerk
plan.md
→ ClerkProvider and a secure token cache
→ no authClient — Clerk's hooks are the client
src/hooks/auth/useAuthedFetch.ts
(bearer token to your backend)
add stripe --replace

Slot in one more technology, or swap a single-select category. Infers what it's replacing from the project itself.

✔ replaced revenuecat → stripe 12 files updated, 3 removed
upgrade

Refresh rules, prompts and docs to the current version using the selection the project already has.

✔ 4 updated · 31 already current ℹ 2 newer AI tools available
review

Static, AI-oriented pass over a generated project across architecture, security, performance and DX.

✖ .env exists but is not gitignored ! eslint-disable at src/lib/analytics.ts:12
analyze

Works on any repo, generated or not — infers the stack from dependencies, and names its evidence.

◆ Next.js — high confidence Architecture 70/100
doctor --for startup-mvp

Checks this machine before you spend a wizard run finding out the hard way.

✔ Node 20.11 ✔ Git ✔ npm ! Xcode not found (informational)

Written for four jobs

Ship it this weekend

One preset, one command, and the assistant already knows your stack. No afternoon spent writing rules files by hand.

Standardise a team

Everyone's assistant reads the same rules whether they're on Cursor, Cline or Copilot. Check the config into git and it stays that way.

Spin up client repos

Same scaffolding every time, deterministic output, and a cost estimate you can paste into a proposal.

Stop hand-writing rules

You already wrote a .cursorrules once. This generates the equivalent for eight tools, per technology, and upgrades them in place.

All of this is the free tier

Every command here runs on your machine, needs no account, and is MIT-licensed. The one paid flag is --idea.

See what Pro costs →