Most AI app generators produce toy demos. App Forge produces structured, deployable codebases because it forces you through a specification step first — no spec, no code.
A Claude skill that turns a conversational app idea into a complete, deployable AI-native application codebase.
App Forge operates as a two-phase pipeline:
- Wizard Phase — A structured conversation that assembles a YAML app specification from your idea
- Generation Phase — Compiles the spec into a multi-file codebase using a stack-specific adapter
| Stack | Frontend | Backend | AI Integration | Deploy |
|---|---|---|---|---|
| Next.js | React + shadcn/ui + Tailwind | App Router + Server Actions | Vercel AI SDK | Vercel |
| FastAPI | Jinja2 + HTMX + Tailwind CDN | FastAPI + Pydantic | Direct SDK calls | Railway / Render |
The spec schema supports six composable interaction patterns, assigned per-route:
- conversational — Chat interface with streaming AI responses
- single_shot — Form → AI processes → result
- batch_process — File upload → async AI processing
- observer — Data dashboard, read-heavy, minimal AI
- autonomous — Background AI process (cron/webhook), no UI
- copilot — AI assists with explicit human approval gates
Trigger the skill in Claude with phrases like:
- "Build me an app"
- "App forge — I want to build a customer support chatbot"
- "Generate an AI app that extracts data from invoices"
- "I have an app idea"
The wizard will walk you through defining your app's routes, agents, data layer, and auth — then generate the full codebase.
app-forge/
├── SKILL.md # Wizard flow + generation orchestration
└── references/
├── spec-schema.md # Canonical YAML schema definition
├── adapter-nextjs.md # Next.js generation constraints
├── adapter-fastapi.md # FastAPI generation constraints
├── failure-modes.md # 17 anti-patterns as negative constraints
└── examples/
├── support-bot.yaml # Conversational + tools + auth example
└── invoice-extractor.yaml # Batch process + structured output example
Drop the app-forge/ folder into your Claude skills directory.
The YAML spec captures everything needed to generate an AI-native app:
meta: # App name, description, stack choice
data: # Database provider + entity models
auth: # Auth toggle + OAuth methods
agents: # AI agents with prompts, models, tools, memory, structured output
routes: # URL paths mapped to interaction patterns + agents
env_vars: # Environment variables (auto-derived)
error_behavior: # How the app handles AI failuresSee references/spec-schema.md for the full schema definition with field-level documentation.
Active and in use. The skill is functional for both Next.js and FastAPI stacks. Failure modes documentation covers 17 known anti-patterns to prevent common generation mistakes.
MIT