PRD Pilot

AI workspace for product managers to turn vague ideas into Requirement Specs, PRDs, demo-ready HTML prototypes, consistency reports, and targeted iteration plans.

中文文档

Demo Flow

What It Solves

PRD Pilot is built for product managers, indie builders, and small teams that need to move from a vague idea to a reviewable solution quickly.

Most generators can output a PRD or a mockup, but the workflow usually breaks when:

• the PRD and demo drift apart
• key pages or flows go missing
• feedback turns into a full rewrite

PRD Pilot keeps one shared Requirement Spec through generation, validation, and iteration so the output stays aligned.

Preview

Requirement Spec	Demo Preview	Consistency Check

PRD Draft	Targeted Iteration	Home Workspace

Core Workflow

flowchart LR
    A["Input idea"] --> B["Structure Requirement Spec"]
    B --> C["Generate PRD"]
    B --> D["Generate Demo plan + HTML"]
    B --> E["Generate Prototype Outline"]
    D --> F["Demo quality gate"]
    C --> G["Consistency Check"]
    E --> G
    F --> G
    G --> H["Targeted Iteration"]
    H --> B

Loading

Key Features

Shared Requirement Spec

PRD Pilot first structures user input into a single internal spec:

• product_name
• product_type
• target_users
• user_pain_points
• core_scenarios
• key_features
• primary_pages
• user_flow
• style_preference
• constraints
• success_criteria

This spec becomes the source of truth for generation, checks, and iteration.

PRD + Demo + Prototype Outline

• PRD: Chinese Markdown draft for review
• Demo: single-file HTML prototype for preview and download
• Prototype Outline: structure, flow, and validation goals

Demo Quality Gate

The demo pipeline validates:

• HTML completeness
• key button presence
• interaction signals
• main page connectivity
• result and feedback state coverage

If quality fails, PRD Pilot attempts one repair pass before returning a structured error.

Consistency Check v2

Built-in checks cover:

• page coverage
• feature coverage
• flow connectivity
• naming consistency
• prototype alignment
• scenario coverage

Reports include:

• severity
• evidence
• issue list
• repair actions that can be mapped into targeted iteration

Targeted Iteration

Instead of regenerating everything, PRD Pilot supports scoped updates such as:

• add page
• modify user
• remove feature
• adjust layout
• change style
• improve data density
• simplify PRD
• clarify flow

Each iteration returns:

• change_summary
• changed_sections
• affected_pages

External Integrations

PRD Pilot now officially includes:

• thin MCP service in mcp/
• Claude Code skill in .claude/skills/prd-pilot/SKILL.md
• shared use_cases layer used by Web API, MCP, and tests

See docs/integration.md for setup details.

Examples

Standard cases are included in prd-pilot/docs/examples/:

• Campus Secondhand Marketplace
• AI Resume Optimizer

Quick Start

1. Clone and start the backend

git clone https://github.com/156181679-dev/prd-pilot.git
cd prd-pilot/prd-pilot/backend
pip install -r requirements.txt
copy .env.example .env
python main.py

Example .env:

OPENAI_PROVIDER=deepseek
OPENAI_API_KEY=your_deepseek_api_key_here
OPENAI_BASE_URL=https://api.deepseek.com/v1
OPENAI_MODEL=deepseek-chat
OPENAI_MAX_TOKENS=0
APP_HOST=127.0.0.1
APP_PORT=8000

2. Start the frontend

cd ../frontend
npm install
npm run dev

The frontend reads VITE_API_BASE_URL if you want to override the default /api path. In local dev, Vite proxies /api to http://127.0.0.1:8000.

3. Open the app

• Frontend: http://127.0.0.1:5173
• Backend health: http://127.0.0.1:8000/api/health
• API docs: http://127.0.0.1:8000/docs

API

Base URL: http://127.0.0.1:8000

Endpoint	Method	Description
/api/health	GET	Health check
/api/model-options	GET	List available models
/api/test-model-config	POST	Test API key + model connection
/api/structure-requirement	POST	Generate Requirement Spec from brief
/api/generate-prd	POST	Generate PRD from Requirement Spec
/api/generate-demo	POST	Generate HTML Demo from Requirement Spec
/api/check-consistency	POST	Cross-check PRD + Demo against Spec
/api/iterate-prd	POST	Scoped PRD update
/api/iterate-demo	POST	Scoped Demo update

generate-demo and iterate-demo return:

• demo_quality
• generation_meta

Structured demo-stage failures return:

• error_code
• stage
• retryable
• detail

Tests and CI

The repository includes:

• backend pytest coverage for use_cases, structured demo errors, and consistency output
• Playwright smoke coverage for the web happy path and demo timeout path
• GitHub Actions workflow for backend tests, frontend build, and browser smoke tests

Tech Stack

Frontend

• Vue 3
• Vite
• Element Plus
• Tailwind CSS
• MarkdownIt
• VueUse
• Playwright

Backend

• FastAPI
• OpenAI-compatible API client
• Pydantic
• Python Dotenv
• FastMCP / MCP
• Pytest

Project Structure

.
+-- prd-pilot/
|   +-- backend/
|   |   +-- main.py
|   |   +-- requirements.txt
|   |   +-- services/
|   |   |   +-- llm_service.py
|   |   |   +-- use_cases.py
|   |   |   +-- mock_llm_service.py
|   |   +-- tests/
|   +-- frontend/
|   |   +-- src/
|   |   |   +-- App.vue
|   |   +-- tests/e2e/
|   |   +-- package.json
|   |   +-- vite.config.js
+-- mcp/
+-- docs/
|   +-- README.zh-CN.md
|   +-- integration.md
|   +-- screenshots/
+-- .claude/skills/prd-pilot/
+-- .github/workflows/
+-- README.md
+-- CONTRIBUTING.md
+-- ROADMAP.md
+-- LICENSE

Documentation

• 中文文档
• Integration Guide
• Contributing Guide
• Roadmap

Current Scope

• prototype output is HTML Demo + Prototype Outline
• consistency checks are rule-based, not AI-score-driven
• targeted iteration is scoped, but not yet a persistent version system
• no image-based prototype generation

License

MIT