Product Introduction
1. Definition: agency-agents is an open-source, MIT-licensed repository that functions as a large-scale AI agent persona library and multi-agent orchestration framework. It provides a comprehensive suite of specialized AI agents engineered to cover every operational role inside a modern digital agency — from frontend development and backend architecture to paid media buying, sales engineering, and community management. The project is distributed as structured Markdown-based agent definitions organized into logical divisions (engineering, design, marketing, paid-media, sales, finance, security, game-development, and more), each encoding an agent's identity, personality traits, core mission, execution workflows, technical deliverables with code examples, and measurable success metrics. Technically, the repository behaves as a prompt-engineering and agent-configuration management system: it ships with an interactive installer (scripts/install.sh), a format converter (scripts/convert.sh), and machine-readable manifests (divisions.json, tools.json) that allow the same agent definitions to be translated into compatible formats for multiple AI coding tools.
2. Core Value Proposition: The value proposition is best summarized as a "coordinated AI workforce for digital agencies" — a complete roster of expert agents in a single repository, ready to deploy. agency-agents directly solves the problem of fragmented, single-purpose AI tools by unifying dozens of specialized, personality-driven AI personas under one installable framework. Instead of relying on generic prompt templates that fail under real-world pressure, teams gain access to battle-tested, deliverable-focused AI agents compatible with Claude Code, Cursor, Codex, Gemini CLI, OpenCode, GitHub Copilot, Aider, Windsurf, Kimi Code, Osaurus, and more. The project was born from a Reddit thread and refined over months of iteration, and it now serves as a primary reference for team leaders, developers, and marketers who want a ready-to-deploy, multi-skilled AI workforce to automate and enhance development, marketing, and community management workflows.
Main Features
1. Cross-Platform Native Desktop App (Agency Agents App): The fastest distribution channel is a native desktop application for macOS, Linux, and Windows. The app browses the entire agent roster and installs selected agents into Claude Code, Cursor, Codex, Gemini CLI, OpenCode, Qwen, Osaurus, and other environments with a single click — no repository cloning and no shell scripting required. The app auto-updates, ensuring that installed agent definitions stay synchronized with upstream improvements. On macOS, it is available via Homebrew Cask (brew install --cask msitarzewski/agency-agents/agency-agents), and precompiled binaries are distributed through GitHub Releases. This lowers the technical barrier for non-developer roles such as marketing managers and paid media strategists who want to activate specialist agents without touching a terminal.
2. Universal Multi-Tool Integration Layer (scripts/install.sh + scripts/convert.sh): For command-line users, agency-agents provides a script-based integration engine. The convert.sh script generates integration files for all supported tools in one pass; the install.sh script then installs agents interactively while auto-detecting which tools are already present on the host machine. Explicit tool-targeting flags are available for: Claude Code, GitHub Copilot, Antigravity, Gemini CLI, OpenCode, OpenClaw, Cursor, Aider, Windsurf, Kimi Code, Codex, Osaurus, Hermes, and Mistral Vibe. The installer also supports advanced selection flags such as --division engineering,security to install entire teams, --agent frontend-developer,ui-designer to install individual agents, --list teams to enumerate all available agent groups, and --dry-run for pre-flight validation without modifying the filesystem. Notably, the installer detects when a selection would exceed the OpenCode runtime's ~119-agent registration limit (an upstream bug that silently drops excess agents) and warns the user to install a subset instead.
3. Division-Based Agent Roster with 20+ Domain Specialties: The repository is organized into modular divisions that mirror a full-service agency structure: academic, design, engineering, examples, finance, game-development, gis, healthcare, integrations, marketing, paid-media, product, project-management, sales, scripts, security, spatial-computing, specialized, strategy, support, and testing. The engineering division alone contains more than 50 specialist agents, including a Frontend Developer (React/Vue/Angular, UI implementation, Core Web Vitals optimization), Backend Architect (API design, microservices), Solidity Smart Contract Engineer (EVM contracts, gas optimization, DeFi), RAG Pipeline Engineer (chunking, hybrid search, re-ranking), Section 508 Accessibility Specialist (WCAG, VPAT/ACR authoring), Rust Refactoring Specialist, LLM Post-Training Engineer (SFT/DPO/GRPO/RLVR), Embedded Firmware Engineer (ESP32/STM32/Nordic, RTOS), and a GaussDB Expert Engineer for Huawei's OLTP database. The marketing division includes a Reddit Community Builder, TikTok Strategist, Xiaohongshu Specialist, WeChat Official Account Manager, Zhihu Strategist, and an X/Twitter Intelligence Analyst. The paid-media division provides a PPC Campaign Strategist, Search Query Analyst, Paid Media Auditor (200+ point account audits), and a Programmatic & Display Buyer. The sales division includes MEDDPICC-based Deal Strategist, Signal-based Outbound Strategist, and Technical Sales Engineer with battlecards. Overall, the roster exceeds 100 agents, surpassing the registration capacity of several underlying runtimes.
4. Production-Ready Agent Persona Architecture: Each agent file in agency-agents is a self-contained operational artifact, not a superficial system prompt. A typical agent definition includes: identity and personality traits (unique voice, communication style, and behavioral approach); core mission and scope boundaries; step-by-step execution workflows; technical deliverables with concrete code examples and implementation patterns; and objective success metrics. For example, the UI Finish-Gate Reviewer is designed to catch "interchangeable" AI-generated UI before ship via evidence-based review and a written design contract; the Autonomous Optimization Architect handles LLM routing, cost optimization, and shadow testing for intelligent API selection; and the Minimal Change Engineer enforces minimum-viable diffs to prevent scope creep. This architecture makes every agent auditable, maintainable, and adaptable — users can treat agents as a reference library, copy them into their own workflows, or install them directly into AI tools.
5. Selective Deployment & Sub-Roster Management: Agencies and individual developers rarely need every specialist at once. agency-agents implements a granular selection system that keeps runtime footprints small and context windows clean. Users can install whole divisions (e.g., --division engineering,security), a single agent (e.g., --agent frontend-developer), or a curated combination of agents across divisions. The interactive wizard walks users through tool detection and team selection, while the --dry-run flag previews exact file operations before execution. This selectivity is particularly important for tools with hard runtime limits: OpenCode currently registers only ~119 agents and silently drops the rest due to an upstream bug, so the installer actively warns users when a selection would exceed the threshold. This makes agency-agents safe for production use across constrained environments rather than producing massive, partially installed configurations.
6. Open-Source, Community-Driven Ecosystem with Governance Files: agency-agents is distributed under the MIT License, which permits commercial use, modification, and redistribution. The repository is actively governed: it includes a security policy (SECURITY.md), contribution guidelines in both English and Chinese (CONTRIBUTING.md, CONTRIBUTING_zh-CN.md), a .gitattributes file for cross-platform consistency, and a .gitignore for clean installs. The project maintains GitHub Issues, Pull Requests, Discussions, Actions, and Projects — signaling an active maintainer team and an engaged contributor base. The repository's popularity in the open-source community (approximately 140,000 stars and 22,800 forks) indicates sustained interest and a large user ecosystem, while the division manifests (divisions.json and tools.json) enable programmatic discovery and custom tooling around the agent roster.
Problems Solved
1. Pain Point: The products and workflows it targets suffer from a suite of compounding problems: (a) fragmented, single-purpose AI tools that each require separate setup, licensing, and prompt-tuning, resulting in integration overload; (b) generic prompt templates that produce generic output — shallow, error-prone, and lacking domain-specific execution depth; (c) no standardized way to share or version-control agent personas across tools, so teams rebuild the same specialized prompts in Claude Code, Cursor, and Codex separately; (d) production readiness gaps, where AI-generated code, copy, or campaign plans lack the success metrics and QA discipline required for real deployments; and (e) onboarding friction, where new team members must reverse-engineer a teammate's ad hoc prompt setup. agency-agents collapses these into one governed, installable framework, meaning the agent definitions are versioned, documented, and reusable across every major AI coding runtime.
2. Target Audience: The product serves a wide but well-defined set of users: (a) software engineering teams — React developers, backend architects, DevOps engineers, mobile developers, and SREs who want installed AI specialists inside Claude Code, Cursor, or Codex that can handle pipeline automation, code review, incident response, and database optimization; (b) technical marketing and growth professionals — SEO specialists, social media managers, content strategists, ASO managers, and community managers seeking a Reddit Community Builder, TikTok Strategist, or App Store Optimizer without coding the prompt infrastructure themselves; (c) paid media practitioners — PPC campaign strategists, search query analysts, and media buyers who need account audit frameworks, GTM/GA4 tracking logic, and creative-testing protocols delivered as reproducible AI agents; (d) sales and RevOps leaders — outbound strategists, discovery coaches, deal strategists, and pipeline analysts looking for repeatable, coachable AI sales methodology; (e) digital agency owners and project managers — who want a standardized, multi-skilled AI workforce to scale delivery without proportional headcount growth; and (f) AI platform engineers — who use the repository as a reference corpus for prompt engineering, agent orchestration, and multi-agent system design.
3. Use Cases: Concrete scenarios where agency-agents is essential include: (a) a web agency onboarding a new eCommerce client installs the WooCommerce Shopping Cart Engineer, Drupal Performance Engineer, and Database Optimizer agents into Claude Code to rapidly audit and rebuild the storefront, cutting discovery time from days to hours; (b) an engineering team adopting Codex for a cloud migration installs the DevOps Automator, SRE, and FinOps Engineer agents with --division engineering,security to handle CI/CD pipelines, SLO monitoring, and cloud cost allocation in parallel; (c) a B2B SaaS startup preparing an enterprise launch uses the Sales Division — Outbound Strategist, Proposal Strategist, and Sales Engineer — to generate multi-channel sequences, technical demos, and competitive battlecards in a single afternoon; (d) a mobile app publisher uses the App Store Optimizer and Growth Hacker agents to audit metadata, design keyword strategy, and build viral loops before a global release; (e) a healthcare or government contractor uses the Section 508 Accessibility Specialist and USWDS Developer agents to produce accessible UI components and VPAT/ACR documentation for compliance; and (f) an AI platform team debugging LLM routing behavior uses the Autonomous Optimization Architect, RAG Pipeline Engineer, and Prompt Engineer agents as reference implementations for internal agent governance.
Unique Advantages
1. Differentiation: Compared to generic prompt libraries and single-purpose AI tools, agency-agents offers a fundamentally more structured approach. Most prompt marketplaces sell one-off chat templates with no install story, no versioning, and no cross-tool compatibility. agency-agents instead ships a complete product lifecycle: machine-readable manifests (divisions.json, tools.json), a converter that generates per-tool integrations, an installer that auto-detects installed runtimes, a desktop app for non-technical users, and a roster of agents that are deliverable-focused and success-metric-driven. Unlike single-purpose AI tools (e.g., a standalone code generator or a social media scheduler), the repository provides a coordinated team of specialists that share a consistent structure and can be deployed across Claude Code, Cursor, Copilot, Gemini CLI, and 10+ other tools simultaneously. Unlike closed, proprietary agent platforms, agency-agents is MIT-licensed — teams retain full control, can modify any persona, keep agents in their own repository, and run them locally without vendor lock-in. The OpenCode 119-agent limit detection is a further differentiator: it demonstrates a level of runtime-aware engineering that generic installer scripts lack.
2. Key Innovation: The central innovation is the reimagining of AI agents as a portable, installable software artifact — a package-manager model applied to prompt-defined specialists. Instead of agents living only as ephemeral chat context, agency-agents treats them as versioned Markdown artifacts that can be converted into the native format of any supported AI tool (Claude Code, Cursor, Codex, Gemini CLI, OpenCode, Copilot, Aider, Windsurf, Kimi Code, Osaurus, Hermes, Vibe, etc.), installed selectively, and updated over time. The combination of a convert.sh translation layer, an install.sh orchestration layer, a native desktop app for auto-updating distribution, and divisions.json/tools.json manifests allows the same roster to be consumed in three modes: browse (reference), install (activate), or adapt (copy into a custom system). That three-mode consumption model, plus the runtime-limit avoidance logic, is what makes this product genuinely unique in the open-source AI agent ecosystem.
Frequently Asked Questions (FAQ)
1. What is agency-agents and how is it different from a prompt library?
agency-agents is an open-source repository containing hundreds of specialized, production-ready AI agent definitions organized by digital agency division — engineering, design, marketing, paid media, sales, finance, security, and more. Unlike a typical prompt library, each agent file encodes identity, personality, workflows, technical deliverables with code examples, and success metrics, and is packaged with installation tooling (scripts/install.sh, scripts/convert.sh) that installs them into Claude Code, Cursor, Codex, Gemini CLI, OpenCode, GitHub Copilot, and other AI coding tools. A native desktop app for macOS, Linux, and Windows provides one-click installation and auto-updates.
2. Is agency-agents free to use and commercially licensable?
Yes. The entire repository is released under the MIT License, which grants free use, modification, and redistribution — including for commercial purposes. Users can install agents directly, fork the repository, or embed agent definitions into proprietary internal workflows without paying licensing fees. The project also maintains a SECURITY.md policy and bilingual contribution guidelines, so third-party contributions are welcome under a clear governance model.
3. How do I install agency-agents with Claude Code or Cursor?
Three methods are supported. For a desktop experience, download the Agency Agents native app (or run brew install --cask msitarzewski/agency-agents/agency-agents on macOS) and click through the install wizard. For terminal users, run ./scripts/install.sh --tool claude-code to install all agents to the Claude Code directory, or ./scripts/install.sh --tool cursor for Cursor. Alternatively, copy a single division manually (e.g., cp engineering/*.md ~/.claude/agents/) to activate specific agents. Selective installs are also possible: ./scripts/install.sh --tool claude-code --division engineering,security or ./scripts/install.sh --tool cursor --agent frontend-developer,ui-designer.
4. Which AI tools and runtimes are supported by agency-agents?
The project officially supports Claude Code, GitHub Copilot, Antigravity, Gemini CLI, OpenCode, OpenClaw, Cursor, Aider, Windsurf, Kimi Code, Codex, Osaurus, Hermes, and Mistral Vibe. The installer auto-detects which tools are present on your machine. OpenCode users should note that the runtime currently registers only ~119 agents and silently drops the remainder due to an upstream bug; the installer warns users and recommends a subset installation using the --division flag to stay under that limit.
5. Can I install only specific agents or teams for my project?
Yes. The installer supports granular selection: --division engineering,security installs entire teams, while --agent frontend-developer,ui-designer installs only the named agents. Running ./scripts/install.sh --list teams displays every available team with agent counts, and --dry-run shows exactly what would be installed before any file changes. This is particularly useful for large agencies that want a custom AI workforce per client engagement without cluttering the tool runtime.
