TL;DR: Agency Agents (msitarzewski/agency-agents) is a collection of specialized, personality-driven AI agents — "a complete AI agency at your fingertips." It has blown past 140,000 GitHub stars (mid-August 2026) to become arguably the most-starred AI agent project of the year. There's now a native desktop app (macOS/Linux/Windows) that installs agents into Claude Code, Cursor, Codex, Gemini CLI, OpenCode, Qwen, and more with one click. It's free, MIT-licensed, and genuinely well-crafted — with the caveat that quality varies by agent and the roster is a grab-bag, not a coherent product.
Table of Contents
1. What Agency Agents Is (and Isn't)
Agency Agents is a growing collection of meticulously crafted AI agent personalities, born from a Reddit thread and months of iteration. The pitch: instead of generic "you are a helpful assistant" prompts, each agent is a specialized expert with a personality, defined processes, and proven deliverables.
Think of it as assembling your dream team — except the team members are AI specialists who never sleep and always deliver. The roster spans everything from "frontend wizards" to "Reddit community ninjas", from "whimsy injectors" to "reality checkers".
Technically, it's a collection of Markdown agent definitions (identity, mission, workflows, code examples, success metrics) plus shell scripts that install those definitions into whatever AI coding tool you use. It is not a new agent runtime — it's a library of agent personas that plugs into the tools you already have.
2. Why It Exploded Past 140k Stars
As of August 10, 2026, the repo sits at ~140,656 stars — up from roughly 8.9k-class projects in the same niche, and trending at or near the top of GitHub Trending for days. A few reasons it caught fire:
- The agency metaphor works. Everyone understands "hire a team of specialists." It makes agents feel tangible instead of abstract.
- Zero friction to try. One install script, no new infra, works with tools people already pay for (Claude Code, Cursor, Codex).
- An official app. The Agency Agents app (native, auto-updating, on macOS/Linux/Windows) removes the last barrier — no terminal required.
- Community momentum. Born on Reddit, PRs welcome, sponsor-supported — it rode the same "agent skills" wave as Superpowers and the official Claude agent-skills repo (we covered that ecosystem in Agent Skills Ecosystem 2026).
3. Installation: App, Claude Code, and Other Tools
Option 1: The desktop app (recommended)
# macOS
brew install --cask msitarzewski/agency-agents/agency-agentsOr download the latest release from the agency-agents-app repo for Windows/Linux. The app browses the whole roster and installs agents into Claude Code, Cursor, Codex, Gemini CLI, OpenCode, Qwen, and Osaurus — with a click, and it auto-updates.
Option 2: CLI with Claude Code
# Install all agents to your Claude Code directory
./scripts/install.sh --tool claude-code
# Or copy a single category
cp engineering/*.md ~/.claude/agents/Then in any Claude Code session: "Hey Claude, activate Frontend Developer mode and help me build a React component."
Option 3: Other tools
# Generate integration files for all supported tools
./scripts/convert.sh
# Install interactively (auto-detects what you have)
./scripts/install.sh
# Or target a specific tool
./scripts/install.sh --tool gemini-cli
./scripts/install.sh --tool codexSupported targets include GitHub Copilot, Antigravity, Gemini CLI, OpenCode, OpenClaw, Cursor, Aider, Windsurf, Kimi Code, Codex, Osaurus, Hermes, and Mistral Vibe.
4. Inside the Roster: What Kind of Agents You Get
The roster is organized into divisions (categories), each with multiple agents. Highlights:
- Engineering — Frontend Developer, Backend Architect, DevOps Specialist, Code Reviewer ("reality checkers")
- Design & Branding — UI/UX specialist, Logo/Brand agents, "whimsy injectors" for personality
- Marketing & Growth — SEO expert, Content Strategist, Reddit community agents, social media specialists
- Product & Ops — Product Manager, QA Tester, Documentation Writer, Support agents
Each agent file contains a full playbook: identity and personality traits, core mission, workflows, technical deliverables with code examples, success metrics, and communication style. The quality bar is genuinely higher than typical prompt collections — these read like onboarding docs for real hires, not one-paragraph prompts.
Reality check: More agents ≠ more capability. The underlying model still does the thinking; these files shape how it behaves. Expect real gains in consistency and workflow adherence, not magic. Also: some agents overlap heavily, and quality varies across the roster.
5. Pros, Cons & Verdict
Get it if you…
- Use Claude Code, Cursor, or Codex daily and want consistent, specialized behavior
- Like the "team of experts" workflow instead of one generic assistant
- Want zero-friction setup (app or one script)
- Value a free, MIT-licensed, community-driven project
Wait if you…
- Expect a coherent product with guarantees — this is a grab-bag of personas
- Are worried about prompt-injection risk from third-party agent files (review before installing)
- Prefer minimal setups — installing 50 agents can add noise
- Need enterprise support or security review
Verdict: Agency Agents earns its stars. It's the best-packaged agent persona collection of 2026 — the app alone is worth the download. Install a handful of agents for your actual workflow (engineering + code review first), review the files once, and skip the rest of the roster. 8.5 / 10
6. FAQ
Q: Is Agency Agents free?
Yes — the agent collection is MIT-licensed and free. The desktop app is also free. You still pay for the underlying AI tools (Claude Code subscription, Cursor, API usage, etc.).
Q: Does it work with OpenAI Codex?
Yes. The app and CLI both support Codex as an install target, along with Claude Code, Cursor, Gemini CLI, OpenCode, Aider, Windsurf, Kimi Code, and more.
Q: Is it safe to install third-party agent files?
Agent files are Markdown instructions — they're not executable code. The practical risk is prompt injection: a malicious agent file could try to steer the model into harmful actions. Install from the official repo (140k+ stars, highly audited), review files you add, and keep agents scoped to sandboxed environments when possible.
Q: How is this different from Claude's official Agent Skills or Superpowers?
Claude Agent Skills and Superpowers are skill/methodology ecosystems — they teach the model how to do things (procedures, workflows, MCP composition). Agency Agents is a persona library — it defines who the agent is (specialist roles with personalities). They're complementary; many users run both. See our Agent Skills Ecosystem 2026 explainer for the full picture.