One spec, 30 correct formats

The Universal Core is written once. Everything below is a deterministic render of it — no second model call, no chance for two files to contradict each other, and no generic “build kit” that fits nothing properly.

30

platforms

Every one gets the artifact it actually parses

1

with hard ceilings

Rendered under the limit and verified by a test, not by hope

2

obsolete formats avoided

Formats these tools stopped reading — we never emit them

Repo agents

These read files committed to your project. Get the layout wrong and the agent silently ignores everything you wrote.

Claude Codepopular
CLAUDE.md + AGENTS.md

Lives at the repo root, in version control. One long prompt rots the context instead.

ZIP
Cursorpopular
.cursor/rules/*.mdc

Directory format. The single-file .cursorrules is obsolete and should not be generated.

no .cursorrulesZIP
Windsurf
.windsurf/rules/*.md
no .windsurfrulesZIP
GitHub Copilot
.github/copilot-instructions.md

One compact file. Rules scattered across several files never fire.

ZIP
OpenAI Codex CLI
AGENTS.md

Where the AGENTS.md standard originated.

ZIP
Gemini CLI
GEMINI.md + AGENTS.md
ZIP
AWS Kiropopular
.kiro/specs/{slug}/requirements.md · design.md · tasks.md

Spec-first: it writes no code until you approve the plan. The ideal target for us — acceptance criteria in EARS format.

ZIP
Google Antigravity
AGENTS.md
ZIP
Zed
AGENTS.md
ZIP
Cline / Roo Code
.clinerules + AGENTS.md
ZIP
Devin
AGENTS.md + knowledge
ZIP
Amp
AGENTS.md
ZIP
Trae
AGENTS.md
ZIP
Aider
CONVENTIONS.md
ZIP

Prompt-to-app builders

No repo — they take a founding prompt and iterate. Each one digests a different amount, and going over the ceiling gets you truncated without warning.

Lovablepopular
Knowledge Base + founding prompt

Their own docs call the Knowledge Base the single most effective thing you can set up. A PRD buried in the prompt gets forgotten.

ZIP + copy-paste
Bolt.newpopular
6-part founding prompt

Purpose · Features · User Flow · Data Model · UI Style · Extras. Go past 400 words and the model chokes.

400-word ceilingZIP + copy-paste
Base44
Single full-stack prompt
copy-paste
Emergent.sh
Agent-segmented brief (UI / logic / DB / API / QA)

A PM agent validates the requirements — the platform hungriest for structured input.

copy-paste
Replit Agentpopular
replit.md + numbered task list

Do not ask for the whole app in one go; it stops halfway.

ZIP + copy-paste
v0 (Vercel)
Per-screen component prompts + design tokens
copy-paste
Softgen
Founding prompt + iterations
copy-paste
Create.xyz
Founding prompt
copy-paste
Databutton
Prompt + Python backend brief
copy-paste
Tempo
Founding prompt (React)
copy-paste
Trickle
Founding prompt
copy-paste
Vybe
Founding prompt
copy-paste
OnSpace.AI
Founding prompt
copy-paste

Mobile-first

These design screens first, then hand off to engineering. We ship the handoff bridge too, so the repo agent doesn't redo your design.

Rork
React Native + Expo prompt + screen flow
ZIP + copy-paste
FireVibe
Screen list + brand tokens + handoff bridge

We also emit a handoff file so you can move to Cursor or Claude Code after export.

ZIP + copy-paste
FlutterFlow AI
Visual + AI brief (Flutter)
copy-paste

What we deliberately do not generate

Formats these tools have moved on from. Shipping them looks like more value and produces exactly nothing — the tool never reads the file.

  • Cursor — no .cursorrules, we emit .cursor/rules/*.mdc instead
  • Windsurf — no .windsurfrules, we emit .windsurf/rules/*.md instead

Why this is one price, not thirty

The expensive part is the core: the evidence, the counter-evidence, the scope boundary. Rendering that into a platform's file layout is template work, so it costs 5 credits and you are never charged twice for the same platform. Pick as many as you want to try.