Agency Agents — Install 260 AI Specialists Into Your Coding Tool
One command adds a full agency to Claude Code, Cursor, Codex and a dozen more
What is Agency Agents?
260 specialist AI personas you can drop into the coding tool you already use
The whole thing in one look
- How do you want to install?
- Desktop app — click, no terminal
- Script — one command
- Point it at your AI tool — Claude Code, Cursor, Codex…
- Pick divisions, not everything — 260 agents is a lot
- Run it — formats convert themselves
- Say “activate Frontend Developer” — in your normal chat
Four tools install into the current folder — check step 4.
[Agency Agents](https://github.com/msitarzewski/agency-agents) is a collection of **260 AI agent personas** — a frontend developer, a backend architect, a security auditor, a GIS analyst, a Reddit community manager — each written as a markdown file with its own expertise, workflow and voice.
They are not a chatbot or an app you talk to. They install **into the tool you already use**, and then you switch one on by asking for it in normal conversation.
- **Works with 15 tools** — Claude Code, Cursor, Codex, Copilot, Gemini CLI and more
- **Install all 260 or just one division** — you pick
- **Free and MIT licensed**, 148K+ stars
- **No API keys, no account** — these are files, not a service
It costs nothing extra to run An agent is a prompt file, so it uses whatever subscription or API key your tool already has. Installing 260 of them does **not** cost 260× — only the one you activate is ever loaded into a conversation.
Before you start
- At least one supported AI coding tool installed (Claude Code, Cursor, Codex…)
- Desktop app route: nothing else — macOS, Windows or Linux
- Script route: git and bash
- Windows on the script route: Git Bash or WSL (the installer is a shell script)
- ~40MB of disk for the clone
Two routes in
The app is genuinely easier — the script gives you finer control
| Desktop app | Script | |
|---|---|---|
| Terminal needed | No | Yes |
| Clone the repo | No | Yes |
| Browse agents before installing | Yes, visually | `--list agents` |
| Pick individual agents | Yes | Yes, `--agent` |
| Updates | Auto-updates itself | `git pull` (see the symlink tip) |
| Scriptable / CI | No | Yes |
Not sure? Take the app If you just want the agents and don't care how they got there, use the desktop app and skip steps 2–4 entirely.
Or let an AI agent do it
Paste this into Claude Code or Codex and it installs the rest for you
There is a pleasing circularity to having an AI agent install your AI agents. This prompt makes it check what you have, show you the roster, and dry-run before writing anything.
Install the Agency Agents collection into my AI coding tool.
Repo: https://github.com/msitarzewski/agency-agents
Do this:
1. Tell me which of these you can see installed before you touch anything:
Claude Code, GitHub Copilot, Cursor, Codex, Gemini CLI, Antigravity,
OpenCode, Qwen Code, Kimi Code, Osaurus, Aider, Windsurf, OpenClaw,
Hermes, Mistral Vibe.
2. Clone the repo shallowly somewhere sensible and cd into it:
git clone --depth 1 https://github.com/msitarzewski/agency-agents.git
3. Show me the roster before installing anything:
./scripts/install.sh --list teams
4. Ask me which divisions I actually want. Do not install all 260 agents
unless I say so.
5. Dry-run the exact command first and show me the plan:
./scripts/install.sh --tool <tool> --division <divisions> --dry-run
6. If the plan looks right, run it for real without --dry-run.
7. Note that cursor, opencode, aider and windsurf install into the CURRENT
directory, not my home folder. If I picked one of those, ask me which
project directory to run it from and cd there first.
8. When it finishes, list the files that were actually written and tell me the
exact sentence I type to activate one agent.
Rules:
- Show me each command before you run it.
- Do not modify anything outside the clone and the target tool's agent folder.
- If a step fails, show me the real error and your fix before continuing.1. Install the desktop app
The no-terminal route — skip to step 2 if you'd rather use the script
[agencyagents.app](https://agencyagents.app) is a native app that browses the whole roster and installs agents into your tools with a click. It keeps itself updated, so new agents show up on their own.
brew install --cask msitarzewski/agency-agents/agency-agentsNot on Homebrew? Grab an installer from the [latest release](https://github.com/msitarzewski/agency-agents-app/releases/latest) — `.dmg` for macOS, `-setup.exe` for Windows, `.AppImage` / `.deb` / `.rpm` for Linux. Apple Silicon and Intel, x64 and ARM builds are all there, so take the one matching your machine.
That's the whole app route Open it, tick the agents you want, choose your tool, click install. Then jump to **step 5** to switch one on.
2. Clone the repo
Script route starts here
git clone --depth 1 https://github.com/msitarzewski/agency-agents.git
cd agency-agentsWindows: use Git Bash or WSL `install.sh` is a bash script. It runs fine on Windows, but through Git Bash or WSL — not PowerShell and not `cmd`. If you'd rather not, the desktop app in step 1 is a native Windows build.
3. Look before you install
See the roster and dry-run the plan
260 agents is more than anyone needs. Two commands let you see what you're choosing from before anything is written to disk.
./scripts/install.sh --list teams
./scripts/install.sh --list agents| Division | Agents | What's in it |
|---|---|---|
| `engineering` | 59 | Frontend, backend, mobile, DevOps, SRE, code review, databases |
| `specialized` | 58 | The odd jobs — orchestrators, accounts payable, identity, trust |
| `marketing` | 36 | Content, growth, social platforms, community |
| `gis` | 13 | Mapping, spatial analysis, geodata |
| `security` | 12 | Auditing, threat modelling, incident response |
| `design` | 10 | UI, UX research, brand, visual storytelling |
| `testing` | 9 | QA, automation, performance, test strategy |
| `sales` | 9 | Outbound, CRM, pipeline, proposals |
| `project-management` | 7 | Planning, shipping, coordination |
| `paid-media` | 7 | Ads, bidding, campaign management |
| `support` | 6 | Customer support, docs, escalation |
| `spatial-computing` | 6 | AR/VR/XR |
| `game-development` | 6 | Gameplay, engines, level design |
| `academic` | 6 | Research writing, citations, peer review |
| `product` | 5 | Discovery, roadmaps, prioritisation |
| `finance` | 5 | Modelling, accounting, analysis |
| `healthcare` | 3 | Clinical and health-domain work |
| `strategy` | 2 | Positioning and decision support |
| `research` | 1 | General research |
./scripts/install.sh --tool claude-code --division engineering,security --dry-run4. Install
One command — it converts formats for you
Run it bare and you get an interactive wizard that detects which tools you actually have and lets you tick boxes:
./scripts/install.shOr say exactly what you want and skip the wizard:
# one tool, two divisions
./scripts/install.sh --tool claude-code --division engineering,security
# named agents only
./scripts/install.sh --tool cursor --agent frontend-developer,ui-designer
# everything, no prompts
./scripts/install.sh --no-interactiveYou don't need to run convert.sh first The README lists `./scripts/convert.sh` as step 1. You can skip it — `install.sh` runs the conversion itself whenever a tool's integration files are missing. Pass `--no-convert` only if you want to stop it doing that.
Four tools install into the folder you're standing in **Cursor**, **OpenCode**, **Aider** and **Windsurf** write into the *current directory*, not your home folder. Run the installer from inside the project you want the agents in — if you run it from the clone, that is where they land, and your real project won't see them.
| Tool | Format | Installs to |
|---|---|---|
| Claude Code | `.md` agents | `~/.claude/agents/` |
| GitHub Copilot | `.md` agents | `~/.github/agents/` + `~/.copilot/agents/` |
| Gemini CLI | `.md` agents | `~/.gemini/agents/` |
| Antigravity | `SKILL.md` | `~/.gemini/config/skills/` |
| Codex | TOML agents | `~/.codex/agents/` |
| Qwen Code | `.md` SubAgents | `~/.qwen/agents/` |
| Kimi Code | YAML specs | `~/.config/kimi/agents/` |
| Osaurus | `SKILL.md` skills | `~/.osaurus/skills/` |
| OpenClaw | SOUL + AGENTS + IDENTITY | `~/.openclaw/agency-agents/` |
| Hermes | lazy-router plugin | `~/.hermes/plugins/` |
| Mistral Vibe | agents + prompts | `~/.vibe/agents/` |
| **Cursor** | `.mdc` rules | **`.cursor/rules/`** — current folder |
| **OpenCode** | `.md` agents | **`.opencode/agents/`** — current folder |
| **Aider** | one `CONVENTIONS.md` | **`./CONVENTIONS.md`** — current folder |
| **Windsurf** | one `.windsurfrules` | **`./.windsurfrules`** — current folder |
5. Switch an agent on
You just ask for it
There is no command for this. Open your tool as normal and say the agent's name:
Hey Claude, activate Frontend Developer mode and help me
build a React dashboard with chartsSwitching is the same — *"activate Backend Architect"* — and the new persona takes over. In Cursor and Windsurf the rules apply to the project automatically, so there is nothing to say at all.
One agent at a time These personas are opinionated on purpose. Running two at once gives you the average of two specialists, which is a generalist — the thing you installed them to avoid.
Worth knowing
Symlink instead of copy, and updates come free `--link` symlinks the agent files rather than copying them. After that a `git pull` in the clone updates every installed agent in place, with no reinstall. The trade-off: don't delete or move the clone.
./scripts/install.sh --tool claude-code --division engineering --linkOpenCode silently drops agents past ~119 This is an upstream OpenCode bug, not a problem with the repo — its runtime registers about 119 agents and quietly ignores the rest. Install a subset with `--division` to stay under the ceiling. The installer warns you when your selection would exceed it.
Install into more than one tool Nothing stops you running the installer once per tool. The same agents in Claude Code and Cursor is a normal setup — they don't conflict, because each tool reads its own folder.
Do I need to clone the repo if I use the app?
No. The app carries the roster itself and installs from it — that's the main reason to use it.
Does this cost anything to run?
No. The agents are markdown files under an MIT licence, and they run on whatever subscription or API key your coding tool already uses. There is no account and no key to add.
Will it overwrite agents I wrote myself?
It writes into the shared agent folder, so a file of yours with the same name as one of theirs would be replaced. Run with --dry-run first and read the list of files it plans to write.
I installed for Cursor and Cursor can't see them
You almost certainly ran the installer from the clone. Cursor reads .cursor/rules/ in the project you have open — cd into that project and run it again from there.
Do I have to install all 260?
No, and you shouldn't. Use --division engineering,security or --agent with specific names. Fewer, more relevant agents are easier to remember and pick from.
How do I uninstall?
Delete the files from the tool's agent folder — the table in step 4 says where each one lands. There is no uninstaller because there is nothing running; they're just files.
Repo: **[github.com/msitarzewski/agency-agents](https://github.com/msitarzewski/agency-agents)** · App: **[agencyagents.app](https://agencyagents.app)** More free tools and setup guides at [rexxulabs.com](https://rexxulabs.com/) — follow [@rexxu.ai](https://www.instagram.com/rexxu.ai/) for the next one.