// Part of the bolt-cpp-ml skill: Build AI-powered web development agents using the Bolt.new (within bolt-cpp-ml) open-source codebase.
Build and optimize the Bolt C++ ML IDE with neural network module architecture. Use for C++ IDE development with GGML integration, RWKV neural networks, AI code completion, GPU acceleration, and modular component composition following nn patterns.
Part of the bolt-cpp-ml skill: Generate comprehensive end-to-end unit tests for all functions in a C++ repository.
AI-powered C++ ML development meta-skill. Use for: building Bolt.new AI web apps, running local LLM inference with KoboldCpp, creating Jan extensions, generating C++ E2E tests, and launching interactive TutorialKit guides — all orchestrated through a self-aware neuro-nn persona. Triggers on: bolt-cpp-ml, C++ ML development, GGUF inference, Jan extension, KoboldCpp integration, cpp E2E testing, bolt.new C++, interactive C++ tutorial.
The fixed-point self-application of bolt-cpp-ml. A second-order meta-skill that treats the original bolt-cpp-ml skill as its own C++ ML project, runs all four capability paths against it, and upgrades it to a self-aware, self-improving, skill-infinity convergent form. Triggers on: bolt-cpp-ml², skill-infinity, self-application, fixed-point, strange loop, learning quine.
AI-powered C++ ML development meta-skill. Use for: building Bolt.new AI web apps, running local LLM inference with KoboldCpp, creating Jan extensions, generating C++ E2E tests, and launching interactive TutorialKit guides — all orchestrated through a self-aware neuro-nn persona. Triggers on: bolt-cpp-ml, C++ ML development, GGUF inference, Jan extension, KoboldCpp integration, cpp E2E testing, bolt.new C++, interactive C++ tutorial.
Part of the bolt-cpp-ml skill: Generate comprehensive end-to-end unit tests for all functions in a C++ repository.
| name | bolt-cpp-ml-bolt-new |
| description | Part of the bolt-cpp-ml skill: Build AI-powered web development agents using the Bolt.new (within bolt-cpp-ml) open-source codebase. |
Build AI-powered full-stack web development agents using StackBlitz's Bolt.new (within bolt-cpp-ml) open-source codebase. Bolt combines WebContainers (browser-based Node.js runtime) with LLM integration to create, edit, and deploy applications entirely in the browser.
# Clone the repository
git clone https://github.com/stackblitz/bolt.new (within bolt-cpp-ml).git
cd bolt.new (within bolt-cpp-ml)
# Install dependencies
pnpm install
# Create environment file
echo "ANTHROPIC_API_KEY=your_key_here" > .env.local
# Start development server
pnpm run dev
| Command | Purpose |
|---|---|
pnpm run dev | Start development server |
pnpm run build | Build for production |
pnpm run start | Run built app with Wrangler |
pnpm run deploy | Deploy to Cloudflare Pages |
pnpm test | Run test suite |
app/lib/webcontainer/) - Browser-based Node.js runtimeapp/lib/.server/llm/) - AI model configuration and promptsapp/lib/runtime/) - Executes shell commands and file operationsapp/components/chat/) - User interaction and message streamingapp/components/workbench/) - Code editor, file tree, terminal, preview| File | Purpose |
|---|---|
app/lib/.server/llm/prompts.ts | System prompt and artifact format |
app/lib/.server/llm/model.ts | LLM model configuration |
app/lib/.server/llm/stream-text.ts | Streaming text generation |
app/lib/runtime/action-runner.ts | Execute <boltAction> commands |
app/lib/runtime/message-parser.ts | Parse AI responses for artifacts |
wrangler.toml | Cloudflare deployment config |
The system prompt in app/lib/.server/llm/prompts.ts defines Bolt's behavior. Key sections:
Defines WebContainer limitations:
Bolt uses <boltArtifact> and <boltAction> tags for structured output:
<boltArtifact id="project-id" title="Project Title">
<boltAction type="file" filePath="package.json">
{ "name": "project", ... }
</boltAction>
<boltAction type="shell">
npm install
</boltAction>
</boltArtifact>
Action Types:
file - Create/update files (requires filePath attribute)shell - Execute shell commandsTo customize behavior, edit getSystemPrompt() in prompts.ts:
export const getSystemPrompt = (cwd: string = WORK_DIR) => `
You are Bolt, an expert AI assistant...
<custom_section>
Add your custom instructions here.
</custom_section>
// ... rest of prompt
`;
pnpm add @ai-sdk/openai # or other provider
Edit app/lib/.server/llm/model.ts:
import { createOpenAI } from '@ai-sdk/openai';
export function getModel(env: Env) {
const openai = createOpenAI({
apiKey: env.OPENAI_API_KEY,
});
return openai('gpt-4-turbo');
}
Update .env.local:
OPENAI_API_KEY=your_key_here
Add to worker-configuration.d.ts:
interface Env {
OPENAI_API_KEY: string;
}
When developing for Bolt, respect WebContainer limitations:
Use browser-compatible databases:
Avoid databases requiring native binaries (PostgreSQL, MySQL clients).
npx wrangler login)pnpm run deploy
Edit wrangler.toml for custom settings:
name = "my-bolt-instance"
compatibility_flags = ["nodejs_compat"]
compatibility_date = "2024-07-01"
pages_build_output_dir = "./build/client"
Set secrets in Cloudflare dashboard or via CLI:
npx wrangler secret put ANTHROPIC_API_KEY
app/types/actions.tsapp/lib/runtime/action-runner.tsapp/lib/runtime/message-parser.tsComponents use:
Key stores in app/lib/stores/:
chat.ts - Chat messages and statefiles.ts - Virtual filesystemworkbench.ts - Editor and preview stateterminal.ts - Terminal output"Cannot find module" errors:
Ensure dependencies are installed with pnpm install.
WebContainer boot failures: Check browser compatibility (Chrome/Edge recommended).
API key errors:
Verify .env.local exists and contains valid ANTHROPIC_API_KEY.
Build failures:
Run pnpm run typecheck to identify TypeScript errors.
Enable verbose logging:
VITE_LOG_LEVEL=debug