The Agency: A Specialized AI Agent Collection
/agency-agents
The Agency is a free, MIT-licensed collection of AI agent personas — a Frontend Developer, a Backend Architect, a Whimsy Injector — installed as markdown prompt files into Claude Code, Cursor, and other AI tools. Reach for it for narrow, opinionated experts instead of one generalist reinventing its approach every session. Skip it if you'd rather write your own system prompts than inherit someone else's tone and workflow choices.
What is The Agency AI Agent Collection?
The Agency is a collection of AI agent personas that started from a Reddit thread and grew through months of iteration into markdown prompt files organized by division — Engineering, Design, Paid Media, and Sales are documented in the README. Each file defines one persona's identity, workflows, and deliverables, and you install it into Claude Code, Cursor, or another AI tool to activate that specialist instead of steering a generalist assistant.
Specialized AI Agent Personalities and Expertise
- ✓Roster organized into named divisions in the README — Engineering, Design, Paid Media, and Sales — each listing agent name, specialty, and "when to use" guidance in a table.
- ✓Each agent file bundles identity and personality traits, core mission and workflows, technical deliverables with code examples, and success metrics, per the README's "Use as Reference" option.
- ✓scripts/install.sh supports targeted installs by tool and by division or individual agent (--tool, --division, --agent, --list teams, --dry-run), so you don't have to pull every persona into your agent directory.
- ✓scripts/convert.sh generates integration files for tools beyond Claude Code — GitHub Copilot, Antigravity, Gemini CLI, OpenCode, OpenClaw, Cursor, Aider, Windsurf, Kimi Code, Codex, Osaurus, Hermes, and Mistral Vibe are all listed in the README as install targets.
- ✓Deeply specific engineering personas beyond a generic coding-assistant role — a Network Engineer scoped to Cisco IOS/IOS-XE, Juniper Junos, and Palo Alto PAN-OS, and a Solidity Smart Contract Engineer scoped to EVM gas optimization and DeFi.
- ✓A warning from the installer is given if a tool-specific agent limit would be exceeded — as the README points out, OpenCode's runtime is set up to register just ~119 agents, and it then silently discards the others. This represents an upstream bug which the --division flag provides a workaround for.
When to Deploy Specific AI Agents
- •Activating a narrow specialist mid-project — e.g. the Incident Response Commander persona in Claude Code during a live outage instead of steering a generalist through post-mortem structure.
- •Installing just one division, like --division engineering,security, when a team only wants dev-facing agents and not the Paid Media or Sales personas.
- •Standing up the same roster across several AI tools via scripts/convert.sh and scripts/install.sh --tool <name>, instead of hand-writing tool-specific prompt files for each one.
- •Using an agent file as reference material — copying its technical-deliverables and success-metrics sections into your own prompt instead of installing it wholesale.
- •Letting non-engineering functions draw on the same repo, since the Paid Media and Sales divisions ship personas like a PPC Campaign Strategist and a MEDDPICC-focused Deal Strategist alongside the engineering ones.
Getting Started: Installing Agency Agents
The README describes two paths. The GUI path: download the Agency Agents desktop app for macOS, Linux, or Windows from the agency-agents-app releases page, or on macOS run brew install --cask msitarzewski/agency-agents/agency-agents; the app browses the roster and installs into your AI tools with no clone or scripts needed, and auto-updates. The CLI path: clone this repo and run ./scripts/install.sh --tool claude-code to install everything, or copy a single division manually with cp engineering/*.md ~/.claude/agents/. For tools other than Claude Code, run ./scripts/convert.sh first to generate tool-specific integration files, then ./scripts/install.sh --tool <name> (or the interactive wizard with no flags) for Cursor, Codex, Gemini CLI, OpenCode, GitHub Copilot, and the other tools the README lists as supported.
Integrating Agents with Your AI Tools
Once installed into Claude Code, you activate a persona in conversation — the README's own example is "Hey Claude, activate Frontend Developer mode and help me build a React component." To narrow what gets installed, install.sh takes --division (e.g. engineering,security) or --agent flags, --list teams shows every team and its agent count, and --dry-run previews a run without writing files. On OpenCode specifically, keep an install under roughly 119 agents — the runtime silently drops anything past that per the README's noted upstream bug, and the installer will warn you if a selection would exceed it.
Strengths
- ✓Enormous breadth beyond generic coding personas — dedicated agents for niche engineering domains like network config (Cisco/Juniper/Palo Alto), Solidity/DeFi, WeChat Mini Programs, and Feishu/Lark integrations sit next to Design, Paid Media, and Sales divisions in the same repo.
- ✓Installer lets you scope precisely to a tool and a division or individual agent (--tool, --division, --agent), so adopting The Agency doesn't mean dumping the entire roster into your agents folder.
- ✓Desktop app removes the CLI step entirely for people who just want to browse and click-install — no git clone, no shell script, and it auto-updates.
- ✓MIT license, so nothing about using or modifying the agent files is gated.
- ✓Each agent file documents its own deliverables and success metrics, not just a role description, per the "Use as Reference" option in the README — useful even if you never install it, just to read and adapt.
Known Limitations of Agency Agents
- △No total agent count or version number appears anywhere in the README — you only see the real scope of a division by running install.sh --list teams yourself.
- △OpenCode's runtime, which registers just ~119 agents and then silently drops any remaining ones, has an upstream bug documented in the README; consequently, a full install requires the --division flag to keep it under that ceiling.
- △Every agent is a markdown prompt file, not executable code — results depend entirely on how well your AI tool follows the persona's instructions, since The Agency adds no independent verification layer.
- △Auto-update is described only for the desktop app; the README doesn't document an auto-update path for agents installed via the CLI scripts.
- △Install tooling (install.sh, convert.sh) is Shell, so the CLI path assumes a Unix-like shell — Windows users without WSL would lean on the desktop app instead.
Alternatives to The Agency AI Agents
Common Questions About Agency Agents
The Agency is a free, MIT-licensed collection of specialized AI agent personas — markdown prompt files covering Engineering, Design, Paid Media, and Sales — that you install into Claude Code, Cursor, and other supported AI tools so you can activate a narrow expert instead of one generalist assistant.
Install The Agency either through its native desktop app (download from the agency-agents-app releases page, or run brew install --cask msitarzewski/agency-agents/agency-agents on macOS) or by cloning the repo and running ./scripts/install.sh --tool claude-code, per the README's Quick Start.
The README lists Claude Code, Cursor, Codex, Gemini CLI, OpenCode, GitHub Copilot, Antigravity, OpenClaw, Aider, Windsurf, Kimi Code, Osaurus, Hermes, and Mistral Vibe as supported install targets for The Agency's agents.
The Agency's install.sh script takes --division (e.g. engineering,security) and --agent flags to install a subset, plus --list teams to see every team and --dry-run to preview a run before writing any files.
The Agency is released under the MIT license, per the repository's license metadata.
The Agency has a companion native desktop app, also called Agency Agents, for macOS, Linux, and Windows at agencyagents.app — it browses the full roster and installs agents into supported tools with a click, then auto-updates, with a macOS Homebrew cask as an alternative install path.
The problem it solves
Developers running AI coding assistants get one default personality that has to context-switch between frontend work, security review, and incident response inside the same conversation. The Agency solves that by pre-writing separate expert personas — each with its own domain deliverables and success metrics baked into the prompt file — so you pick the specialist for the task (a Database Optimizer for a slow query, an Incident Response Commander for an outage) instead of steering one generalist through every domain shift.
Who should try it — and who should skip
Reach for The Agency if you're already driving Claude Code, Cursor, or another supported tool and want a ready cast of narrow specialists — network config, incident response, Solidity, paid media, sales — instead of steering one generalist through every domain shift. Skip it if your workflow is a single small codebase where the default assistant already covers what you need, or if you'd rather hand-write your own system prompts than adopt someone else's tone and process choices.
Related repositories
Is agency-agents worth your time?
ChatGPT, Claude and Perplexity can all read this page. Ask one of them what it makes of agency-agents.