Agency-Agents: 280+ Persona-Based AI Agents for Solo Developers

One-Line Summary

agency-agents is a massive collection of role-based AI agents: each agent is a standalone markdown file defining identity, personality, mission, workflow, technical deliverables with code examples, success metrics, and communication style — designed to be dropped into Claude Code, Cursor, or 13+ other AI coding tools.

Core Highlights

Highlight 1: 280+ Roles, 18 Divisions, Full Industry Coverage

The repository currently contains 280+ persona agents across 18 divisions (README states 230+, actual file count exceeds 280 and growing). Divisions and representative roles include:

Engineering : Frontend Developer, Backend Architect, SRE, Solidity Contract, RAG Pipeline

Design : UI/UX Designer, Whimsy Injector

Marketing : Xiaohongshu Expert, WeChat Official Account Operator, Zhihu Strategist, Bilibili/Douyin/Kuaishou/Weibo

Sales : Outbound Strategist, Sales Coach

Product/Project Management : Product Manager, Scrum Master

Finance : Financial Analyst, Investment Banking Modeling

Game Development : Level Designer, Numerical Designer

GIS / Spatial Computing : Mapping, Spatial Analysis

Medical / Security / Research : Medical Compliance, Security Audit, Academic Research

The Chinese ecosystem coverage is its strongest differentiator — WeChat Mini Program Developer, Feishu Integration, WeChat Official Account, Xiaohongshu Expert, Zhihu Strategy, Baidu SEO, Cross-border E-commerce, Private Domain Operation (Enterprise WeChat), Government Digitalization (ToG), HR/Recruitment/Study Abroad Advisor — virtually unmatched in similar projects.

Highlight 2: 15+ Tools One-Click Install, Subset Selection

# Install all agents to Claude Code
./scripts/install.sh --tool claude-code

# Install only engineering + security divisions
./scripts/install.sh --tool claude-code --division engineering,security

# Install only specific roles
./scripts/install.sh --tool cursor --agent frontend-developer,ui-designer

install.sh

auto-detects installed tools, supporting Claude Code, Codex, Cursor, Gemini CLI, OpenCode, OpenClaw, Copilot, Antigravity, Aider, Windsurf, Kimi Code, Hermes, Mistral Vibe, DeepSeek Harness, Osaurus, Qwen (15+ total). --division / --agent flags let you install subsets instead of all 280+.

Highlight 3: Native Desktop App + Prebuilt Team Combos

Companion native app agencyagents.app (macOS/Linux/Windows, brew install --cask msitarzewski/agency-agents/agency-agents or download from GitHub releases) lets you browse all roles, click-to-install, and auto-update — lowering the barrier to a single click.

Prebuilt team combos in strategy/runbooks.json — e.g., startup-mvp — install the full crew needed for a startup MVP with one command, no manual picking.

Core Principle & Architectural Differences

Its essence is the persona methodology : not a code framework, but structured markdown capturing what an expert looks like. Each agent file contains four sections: identity & personality, core mission & workflow, technical deliverables with code examples, success metrics & communication style.

Compared to peers, its differentiation is clear:

vs gstack (YC President's suite) : gstack has 23 roles focused on one sprint process (depth); agency-agents has 280+ roles spanning all industries (breadth).

vs generic prompt templates : not a one-liner "you are a frontend dev"; each persona includes deliverables, success metrics, and communication style.

Installation-side engineering is solid: convert.sh generates integration files for 15+ tools in one pass; install.sh auto-detects environment, supports dry-run preview, and --list teams/agents lists available combos.

Installation & Configuration

Method 1 (Recommended): Native App

brew install --cask msitarzewski/agency-agents/agency-agents
# Or download from github.com/msitarzewski/agency-agents-app/releases

Method 2: CLI Scripts

git clone https://github.com/msitarzewski/agency-agents.git
cd agency-agents

./scripts/convert.sh          # Generate integration files for all tools
./scripts/install.sh          # Interactive wizard: pick tool + pick team
./scripts/install.sh --list teams  # Show all teams and agent counts

After install, in Claude Code just say:

"Activate Frontend Developer mode, help me write a React component"

to invoke the persona.

Method 3: As Reference Docs

Each agent file is a high-quality structured document; copy, rewrite, or trim the parts you need.

Team Adoption Recommendations

Pilot : Install --division engineering subset first; let the team run Code Reviewer, SRE, Git Workflow Master roles through a PR gate to validate effectiveness.

Enforce : Codify Code Reviewer and Security Architect checks into the PR pipeline as the first line of AI-side automated review.

Standardize : Write your team's engineering conventions and naming rules as custom agent markdown files alongside official roles, building your own role library.

Promote : Use strategy/runbooks.json or --agents-file to pin the exact crew each project needs; newcomers run --agents-file team.txt for one-click alignment, reducing decision overhead.

Scenario Fit Checklist

Solo dev / one-person company wanting "one person = one company" → Bullseye fit.

Need to code AND run Chinese marketing (Official Account/Xiaohongshu/Zhihu/Bilibili) → Chinese ecosystem is unique.

Developer wanting professional roles for Claude Code/Cursor → 15+ tools fully covered.

Only want a deep, orchestrated sprint process → That's gstack's strength.

Heavy reliance on deep custom agent tuning → Roles are ready-made; deep customization still requires your own edits.

Pros, Cons & Pitfalls

Pros : Extreme coverage (280+ roles, 18 divisions, Chinese ecosystem); MIT free; native app lowers barrier; scripts are production-grade (auto-detect, dry-run, runbooks).

Cons & Pitfalls :

Fundamentally a prompt methodology. Role expertise depends on underlying model capability and your willingness to change workflow to use persona activation.

Quantity itself is a burden. 280+ roles: how to pick, how to digest — real cognitive cost. README notes OpenCode upstream bug registers only ~119 roles; installing too many gets silently dropped. Official advice: install subsets via --division.

Commoditized space. Competes with superpowers / ECC / ponytail / i-have-adhd — all skill/persona methodologies. Moat is massive role count + full industry + Chinese ecosystem, not a unique mechanism.

Primarily solo-maintained. Despite community contributors, core maintenance is concentrated in one person; long-term evolution pace bears watching.

Chinese scenario quality varies. Many Chinese agents exist, but individual role depth needs your own validation; don't assume role existence = expert level.

Closing Thoughts

Hiring a team that never sleeps, never complains, and always delivers is every OPC's dream. But AI can only补足 some capabilities; truly building a dream team isn't that simple. This open-source project is essentially a persona/prompt methodology — its value hinges on the underlying model's ability and whether the user adopts the new workflow. With 300+ roles, selection and digestion carry cognitive cost, so clarifying your own team composition matters more than blindly picking roles.

GitHub Repository: github.com/msitarzewski/agency-agents