The most comprehensive open-source product development skill for Claude.

80 agents · 36 frameworks · Research-first, AI-native, depth to the Mariana Trench · Solo founder Day 0 → IPO → public company

This is a Claude Skill. Claude reads SKILL.md. This README is for human visitors.

Product Architect turns Claude into an entire product organization.

Most AI product advice is one generalist answering everything at surface level. This is the opposite: 80 specialized department heads covering Discovery, PRD, Engineering, Security, Privacy, Pricing, Growth, RevOps, Legal, Finance, Governance and more. Each carries a department-head-depth playbook, routed to your question by a context engine that loads only what's relevant.

Ask "should we build this?" and you don't get an opinion. You get a cited market verdict (does it already exist, who ships it, where's the gap), then a PRD with the edge cases, an architecture, a pricing model, a compliance read, and a launch plan. Each comes from the agent that owns that domain, cross-checked against the others.

What makes it different:

Who it's for: solo founders who need a whole company's worth of expertise · product managers who want PRDs with every edge case · engineering leads needing architecture, security, and test strategy · teams operating at Fortune-500 caliber without Fortune-500 headcount.

Honest scope: this is an educational and operational framework built through human-AI collaboration, not professional advice. Legal, financial, security, and HR content requires qualified review before real-world use. See references/DISCLAIMER.md.

Skills require a paid plan. Free tier does not have access to Skills, including "Add from GitHub."

1. Download this repo as ZIP (Code → Download ZIP)
2. Go to Claude.ai → Settings → Capabilities → Skills
3. Upload the ZIP
4. Test: Ask Claude "Help me build a product" or "Write a PRD for [feature]"

git clone https://github.com/ankitjha67/product-architect.git
# Place in your Claude Code skills directory (~/.claude/skills/)

Use the /v1/skills endpoint with the container.skills parameter. Requires Code Execution Tool beta.

If you're on the free plan, you can still use everything in this repo:

1. Guided mode (recommended) - Open START-HERE.md on GitHub, click Raw, copy everything, paste into a new Claude chat. Describe your idea - Claude will ask what you're building, identify the right agents, give you exact GitHub URLs to paste them one at a time, and guide you through the entire product development process step by step.

2. Projects - Create a Project in Claude.ai, paste SKILL.md + SMART-LOADER.md as project knowledge. Then paste individual agent/framework files as needed per conversation.

3. Direct paste - Open any file on GitHub (e.g. frameworks/founders-playbook.md), copy the raw content, and paste it into a regular Claude chat. Every file works standalone.

4. Claude Code - If you have Claude Code access, clone the repo into your skills directory. Skills load automatically.

Organized by product lifecycle - each operates at department-head depth:

India · US · EU · UK · Southeast Asia (Singapore, Indonesia, Thailand, Vietnam, Philippines, Malaysia)

AI-Native - A modern AI-engineering layer runs through the whole system: the AI Engineering Stack framework (LangGraph orchestration, RAG with hybrid retrieval + reranking + GraphRAG, evals-in-CI, guardrails, observability, and the L0→L5 maturity ladder), plus AI Department Playbooks giving all 80 departments a concrete LLM/RAG/agent application. Defaults to the latest Claude models (Opus 4.8, Sonnet 5, Haiku 4.5) with adaptive thinking, MCP for integrations, and Security (09) + Privacy (39) sign-off on any AI feature touching untrusted input or personal data.

Research-First Depth - Before recommending building anything, agents run the Deep Research Protocol (Agent 47): an end-to-end market investigation that returns a grounded verdict - "this already exists, here are the competitors + citations, refine it" or "this is white-space, no competition/citations in this niche" (with the honest caveat that absence of evidence isn't proof of novelty, plus a "why is it empty?" analysis). Every agent inherits a Depth Rubric (L0 surface → L4 Mariana Trench), must grade itself L3+, and never fabricates a company, statistic, study, patent, or URL.

Smart Loading - SMART-LOADER.md routes each request to only the relevant agents. Scores agents 0-10 on relevance, handles multi-intent requests, supports dynamic mid-conversation loading. Never loads everything at once. Free tier: 3 agents/turn. Pro: 5 agents/turn.

Cross-Agent Governance - 5-level authority hierarchy prevents contradictions: Compliance (override) → Privacy/DPO (override) → Security (override) → Finance (veto) → Chief Reviewer (veto). 5-step conflict protocol: stop, state, apply hierarchy, document, flag.

KDR Memory - Key Decision Records with sequential numbering, conflict detection (scan-before-record), and SUPERSEDED mechanism. Survives chat compaction. Works on free tier. Paste MASTER KDR into new conversations to resume.

Agent Standards - Shared quality protocol inherited by all agents: before/during/after checklists, iterative refinement loops, performance rules, cross-reference table mapping every agent to related frameworks.

product-architect/
├── SKILL.md                     ← Claude reads this (skill entry point)
├── SMART-LOADER.md              ← Context router + memory engine
├── START-HERE.md                ← Free tier entry point (paste into Claude)
├── README.md                    ← You are here (for humans)
├── ENHANCEMENTS.md              ← Enhancement roadmap (what was added & why)
├── CONTRIBUTING.md              ← Conventions, quality bar, wiring checklist
├── CHANGELOG.md                 ← Version history (semver, matches SKILL.md)
├── LICENSE                      ← MIT
├── .github/                     ← CI (validator on every PR), issue & PR templates
├── agents/                      ← 80 agent files (00-79)
├── frameworks/                  ← 36 framework files
├── references/
│   ├── compliance/              ← Country deep-dives (5 files)
│   ├── agent-standards.md       ← Quality protocol for all agents
│   ├── DISCLAIMER.md            ← Professional review requirements
│   ├── github-readme.md         ← Extended documentation
│   └── industry-references.md   ← Best-in-class by function
└── tools/
    ├── navigator.jsx            ← Interactive web UI for founders
    └── validate_repo.py         ← Validator (counts, numbering, links, fences, doc-consistency)

This is an educational framework created through human-AI collaboration, not professional advice. All legal, financial, security, and HR content requires qualified professional review before real-world use. See references/DISCLAIMER.md for full details including AI authorship considerations and human involvement requirements.

MIT - see LICENSE.

Contributions welcome - new agents, deeper frameworks, industry-specific extensions, country compliance additions, bug fixes. See CONTRIBUTING.md for the conventions, quality bar, and wiring checklist, and CHANGELOG.md for version history.

Built through iterative human-AI collaboration using Claude (Anthropic). Compliant with Anthropic's official skill guide. 25/25 skill compliance checks passed.