Imagine having 230 AI specialists at your fingertips – an editorial expert for marketing texts, a security auditor for code reviews, a UX designer for interface feedback, a Reddit community ninja, and a lease contract lawyer. All waiting to be activated by you in Claude Code. No joke.
The Agency is a GitHub repository with over 154,000 stars, born from a simple idea: specialized AI agents for every conceivable task – as Claude Code prompts, organized like a real agency with departments.
The repository was started by msitarzewski after a Reddit thread about specialized AI agents went viral. Since then, over 24,000 forks and hundreds of contributors have built on it. The README claim says it all: "A complete AI agency at your fingertips."
The agents are organized into 18 departments:
| Department | Example Agents |
|---|---|
| Engineering | Architect, Code Reviewer, Tech Lead, Debugging Specialist, Database Designer |
| Design | UI Designer, UX Researcher, Design System Guardian, Accessibility Expert |
| Marketing | Copywriter, SEO Strategist, Social Media Manager, Brand Voice Guardian |
| Sales | Cold Email Specialist, Demo Script Writer, Objection Handler, CRM Optimizer |
| Product | Product Manager (APR), User Story Writer, Roadmap Planner, A/B Test Designer |
| Finance | Financial Analyst, Budget Planner, Invoice Reviewer, Tax Assistant |
| Healthcare | Clinical Trial Writer, ICD Coder, Medical Literature Reviewer, Patient Communicator |
| Security | Penetration Tester, Compliance Auditor, Incident Responder, Threat Modeler |
| Strategy | Competitive Analyst, OKR Coach, Pitch Deck Reviewer, Market Researcher |
| Project Management | Scrum Master, Risk Manager, Stakeholder Communicator, Retro Facilitator |
| Research | Literature Reviewer, Hypothesis Generator, Methodology Advisor, Data Interpreter |
| Support | Triage Agent, Knowledge Base Curator, Escalation Handler, SOP Writer |
| Academic | Paper Reviewer, Grant Writer, Thesis Advisor, Citation Manager |
| Game Development | Level Designer, Narrative Designer, Game Balance Analyst, QA Tester |
| Sales | Email Campaign Specialist |
| Testing | Test Case Generator, Regression Analyzer, Performance Tester, E2E Scenario Writer |
| GIS | Map Analyst, Geodata Expert |
| Spatial Computing | AR/VR Specialist, 3D Model Reviewer |
… and many more. Over 230 specialized agents in total.
Each agent is a clearly defined prompt template with:
Usage is simple:
git clone https://github.com/msitarzewski/agency-agents.git
cp agency-agents/engineering/architect.md ~/.claude/agents/
/claude-code:agent architect
Or you copy individual agent prompts directly from the repository and paste them into your Claude Code conversation.
1. From Reddit thread to movement
What began as a discussion about specialized AI assistants has grown into one of the largest prompt collections in the world. The author himself says: "What started as a Reddit thread about AI agent specialization has grown into something remarkable."
2. Quality through specialization
Instead of a generic "Help me with code," you get an agent that knows exactly what to do. The security auditor thinks like a pentester, the UX designer like a Nielsen Norman graduate, the financial analyst like an auditor.
3. Community-driven
The project thrives on contributors. Anyone can submit new agents, improve existing ones, or report bugs. There are Discussions, Issues, active development.
4. Translations
The agents are available in multiple languages – Korean, Japanese, Vietnamese. German translations are still open (who wants to join?).
Quality varies. With 230+ agents from different authors, quality is not uniform. Some agents are excellently researched, others seem hastily put together. The stars of the repository are the Engineering and Security departments.
Claude Code-centric. The agents are tailored to Claude Code. With other coding agents (Codex CLI, Cursor), they only work with adjustments.
The Agency is more than a prompt collection. It is an operating system overlay for AI-powered work: Instead of having to think anew each time which prompt you need, you rely on proven, community-reviewed templates.
I find particularly valuable:
The Agency is an impressive community project that shows how far specialized AI agents have come in 2026. 154,000 stars don't lie – the idea hits a nerve. Anyone working with Claude Code will certainly find an agent here that improves their own workflow.
Only limitation: Don't trust blindly. Check each agent against your own requirements before first use. But that applies to human employees as well.