AI PRD Generator

Production-ready PRDs from Claude Code

Stop writing PRDs from scratch. One command generates nine verified files — requirements, technical spec, user stories, roadmap, JIRA tickets, test cases, and a claim-by-claim verification report.

Install | What you get | How it works | Commands | Under the hood

---

Install

One-command setup (Claude Code, Cowork, VS Code)

claude plugin marketplace add cdeust/ai-prd-generator-plugin
claude plugin install ai-prd-generator

That's it. The plugin registers its MCP server (7 tools) and 3 commands. Restart Claude Code and you're running.

What gets installed

Component	What it does	Automatic?
MCP Server	7 tools for license validation, config, health checks, strategy listing	Yes
3 Commands	/ai-prd-generator:generate-prd, /ai-prd-generator:index-codebase, /ai-prd-generator:validate-license	Yes
Skill	Full PRD generation workflow with 64 hard output rules	Yes

Alternative install methods

Local development (--plugin-dir)

git clone https://github.com/cdeust/ai-prd-generator-plugin.git
claude --plugin-dir ./ai-prd-generator-plugin

Manual setup

git clone https://github.com/cdeust/ai-prd-generator-plugin.git
cd ai-prd-generator-plugin
./scripts/setup.sh

---

What You Get

One command produces nine files:

File	What's inside
prd-overview.md	Executive summary, goals, scope, and strategic context
prd-requirements.md	Functional and non-functional requirements with traceability
prd-user-stories.md	User stories with acceptance criteria mapped to requirements
prd-technical.md	Technical specification — architecture, data models, API contracts
prd-acceptance.md	Detailed acceptance criteria with test conditions
prd-roadmap.md	Implementation phases, milestones, and dependency ordering
prd-jira.md	Story-pointed epics and tickets, ready to import
prd-tests.md	Unit, integration, and e2e test cases mapped to requirements
prd-verification.md	Claim-by-claim audit — structural integrity, consistency, completeness

See a real example in examples/ — a 2,766-line PRD for a policy-parameterized prompt system, with 24/24 structural integrity checks passed.

---

How It Works

1. You describe what you need — "auth system with OAuth2", "migrate from REST to GraphQL", or paste a Slack thread
2. Clarification loop — The plugin asks structured questions until it reaches 90%+ confidence (or you say "proceed")
3. Strategy selection — Picks from 15 thinking strategies based on your PRD type and complexity
4. Generation + verification — Writes every section, then verifies each claim against 6 independent algorithms
5. Nine files land in your directory — ready for review, handoff, or sprint planning

---

Commands

Command	What it does
/ai-prd-generator:generate-prd	Generate a PRD with interactive clarification
/ai-prd-generator:index-codebase /path	Index a codebase for context-aware generation
/ai-prd-generator:validate-license [key]	Check license tier or activate a license key

Generate with codebase context

/ai-prd-generator:index-codebase /path/to/your/project
/ai-prd-generator:generate-prd

The PRD will reference your actual files, architecture patterns, and dependencies.

---

Under the Hood

PRD types	8 — Feature, Bug, Incident, Proposal, MVP, POC, Release, CI/CD
Thinking strategies	15 — from MIT, Stanford, Harvard, Anthropic, OpenAI, DeepSeek research (2024-2026)
Verification algorithms	6 — adaptive consensus, zero-LLM graph verification, multi-agent debate, atomic claim decomposition
Output rules	64 hard rules enforced per section
Stack support	Python, TypeScript, Go, Rust, Java, Kotlin, Swift

MCP Tools

7 tools reference

Tool	What it does
validate_license	Validate or activate an AIPRD license key
get_license_features	Get available features for current license tier
get_config	Read skill configuration sections
read_skill_config	Read the full skill config
check_health	Health check — environment, version, config status
get_prd_context_info	Get PRD type configuration (sections, clarification range, RAG hops)
list_available_strategies	List available thinking strategies for current tier

License Tiers

Feature	Free	Licensed
PRD types	Feature, Bug	All 8
Thinking strategies	2 (zero-shot, chain-of-thought)	All 15
Verification	Basic	Full multi-judge
Clarification rounds	3 max	Unlimited
Business KPIs	Summary only	Full

---

Part of a Bigger System

This plugin handles PRD generation — Stage 4 of a 10-stage autonomous development pipeline.

ai-architect-feedback-loop is the full pipeline: it takes a research finding through impact analysis, PRD generation, implementation, gate enforcement, verification, and deployment — all autonomous, all verified.

---

System Requirements

Cowork Plugin (Claude Desktop)

• Claude Desktop (macOS or Windows) with a paid plan (Pro, Max, Team, or Enterprise)
• No additional dependencies — the plugin runs inside Cowork's sandboxed VM

CLI Terminal Mode

• macOS, Windows, or Linux
• Node.js 18+
• Claude Code (Anthropic)

---

Troubleshooting

Issue	Solution
Commands not found	Run /reload-plugins or restart Claude Code
MCP server not starting	Verify Node.js 18+ is installed: node --version
Plugin not visible in Cowork	Cowork only runs on Claude Desktop (macOS/Windows)
Clarification loop won't end	Say "proceed" to skip to generation
License key rejected	Check the key starts with AIPRD- and try again

---

Built by Clement Deust