🔥 Forge — AI-Powered Judgment Simulator

Develop real-world judgment through high-stakes AI scenarios.

Quick Start

# 1. Install dependencies
npm install

# 2. Set up environment
cp .env.example .env
# Edit .env and add your ANTHROPIC_API_KEY

# 3. Build the frontend
npm run build

# 4. Run the server
ANTHROPIC_API_KEY=your_key node server.js

Then open http://localhost:3001

Development (hot reload)

Run two terminals:

# Terminal 1: Backend
node server.js

# Terminal 2: Frontend (Vite dev server with proxy)
npm run dev

Frontend runs on http://localhost:5173 and proxies /api calls to the Express server.

Demo Mode

To demo without a real API key:

DEMO_MODE=true node server.js

Or in your .env:

DEMO_MODE=true

Demo mode returns realistic hardcoded responses that feel pressure-filled and real.

Switching AI Providers

All AI calls are isolated in src/lib/ai.js. See the comment block at the top of that file for instructions on switching to OpenAI or Groq — only server.js needs to change on the backend.

Project Structure

forge/
├── server.js              # Express backend (two routes: /api/chat, /api/debrief)
├── src/
│   ├── App.jsx            # Root component, screen routing
│   ├── main.jsx           # React entry point
│   ├── data/
│   │   └── scenarios.js   # All scenario content (add new scenarios here)
│   ├── hooks/
│   │   └── useForge.js    # Central state + API orchestration
│   ├── lib/
│   │   └── ai.js          # All AI/API calls (change this to switch providers)
│   └── screens/
│       ├── HomeScreen.jsx    # Scenario selection
│       ├── ScenarioScreen.jsx # Live chat interface
│       └── DebriefScreen.jsx  # Growth Report visualization
├── index.html
├── vite.config.js
└── package.json

Adding New Scenarios

Edit src/data/scenarios.js and add a new object to the scenarios array. No other files need to change.

Each scenario needs:

• id — unique string
• title, subtitle, tags, durationMin, intensity
• setup — brief situation description shown to user
• characters[] — array of {id, name, role, initial, color}
• systemPrompt — the full system prompt sent to Claude
• openingMessage — the first AI message the user sees

Assumptions Made

1. Vite for frontend build — The spec said "no router needed" so this is a single SPA with state-based screen switching.
2. SSE streaming — /api/chat streams Server-Sent Events. The frontend reads them with the Fetch Streams API (supported in all modern browsers).
3. claude-sonnet-4-5 — Used for both chat and debrief. The spec said claude-sonnet-4-6 which doesn't exist yet; claude-sonnet-4-5 is the latest Sonnet.
4. Opening message baked into scenario data — Rather than having the backend "start" the conversation, the first character message is defined in scenarios.js and rendered immediately without an API call. This keeps the scenario feeling instant.
5. No auth/sessions — This is a single-user demo app with no persistence. Conversation state lives in React state only.