AgenticPod

An agentic product lifecycle accelerator. Give it a product idea, and it runs a staged pipeline across four agent roles (PM, Research, Design, Engineering) to produce structured artifacts — PRD, research plan, prototype spec, tech spec, and backlog.

Claude Code is the UI. No custom CLI. You interact conversationally, and Claude Code drives the pipeline.

Quick Start

1. Install dependencies:

   npm install

2. Set your LLM API key:

   export ANTHROPIC_API_KEY="your-key"

3. Open the project in Claude Code and describe your product idea. Claude Code will use the CLAUDE.md instructions to drive the pipeline.

How It Works

1. You describe a product idea to Claude Code
2. Claude Code initializes a project and creates idea.json
3. Each stage runs an LLM-powered agent that reads prior artifacts and produces new ones
4. Between stages, Claude Code shows you a summary and asks to proceed
5. All artifacts are written to ./artifacts/<project_id>/

Pipeline Stages

Stage	Agent	Output
0: Intake	—	idea.json
1: PM Framing	PM	PRD.md
2: Research	Research	Research.md
3: Design	Design	PrototypeSpec.md, FigmaLink.json
4: Engineering	Engineering	TechSpec.md, Backlog.md
5: Iterate	—	Updated specs, IterationLog.md

Configuration

LLM Provider

Edit config/llm.json to switch between Anthropic and OpenAI:

{
  "provider": "anthropic",
  "model": "claude-sonnet-4-5-20250929",
  "apiKeyEnvVar": "ANTHROPIC_API_KEY"
}

Figma MCP (Optional)

See docs/figma-mcp-setup.md for instructions on connecting Figma for prototype round-trips.

Architecture

src/
  index.ts              # Public API exports
  types/                # TypeScript types and Zod schemas
  providers/            # LLM provider interface (Anthropic, OpenAI)
  agents/               # Agent roles (PM, Research, Design, Engineering)
  conductor/            # Pipeline orchestrator and state management
  artifacts/            # File-based artifact store
  templates/            # Artifact template definitions
  quality/              # Validation and quality checks
  figma/                # Figma MCP config and snapshot sync

Modes

• Mode A (default): Human approves each stage transition via conversation
• Mode B: Auto-advances if quality checks pass (autonomy: 'medium')