Skip to content
Get packs

Agency Agents

A large collection of AI agent prompt files, each set up as a specialist role for tools like Claude Code.

GitHub Download source
Stars
166k
Forks
26.9k
Licence
MIT
Language
Shell
Last update
12 days ago
Type
AI model / agent

What it is

The Agency is a collection of markdown files, each describing an AI agent with a specific role, personality, processes, and deliverables, such as a frontend developer or a Reddit community specialist. You load the agent you need into a tool like Claude Code, Cursor, or Gemini CLI, and the tool then works in that role. The README also describes a native desktop app and shell scripts that install the agents into supported tools.

  • Copy a single agent file into your Claude Code agents folder and activate it by name in a session
  • Install a division such as engineering or security into a supported tool with the install script
  • Install a team of specific agents using the runbook roster files
  • Use an agent file as a template to write your own specialist prompt

Great for

  • Developers who use Claude Code or Cursor and want role-specific prompts for common tasks
  • Small teams wanting a ready-made set of specialist roles to adapt
  • People who want reference prompts to copy and modify for their own agents
  • Users who want to install only one division, such as engineering or security

Not ideal for

  • People looking for a tested, verified tool; the README describes the agents as prompt personalities, so output quality depends on the underlying AI tool and the user's review
  • Users who need guaranteed accuracy in specialist fields like smart contracts, firmware, or network configuration, since the README does not show validation for these outputs
  • Anyone wanting a single program with fixed behavior rather than prompt-based roles
Step-by-step guide: how to use Agency AgentsInstall, first run, examples, tips and FAQs

How to get it

  1. 1

    Clone and build from source

    Clone the repository

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

    Install all agents to Claude Code

    ./scripts/install.sh --tool claude-code
  2. 2

    Run it

    • Clone the repository and open its folder
    • Run ./scripts/install.sh --tool claude-code to install all agents into Claude Code
    • Start a Claude Code session and ask it to activate a named agent, such as Frontend Developer mode
    • Describe a task, such as building a React component, and check the output
  3. 3

    Get the most out of it

    • Use --division or --agent flags to install only the roles you need
    • Run --dry-run first to preview what a selection will install
    • Use --list teams to see every team and its agent count before installing
    • Use the runbook rosters with --agents-file to install a prepared team

Downloads

Source code (.zip)main branch · straight from GitHub

Also on GitHub as mirrors or translations (not listed separately): jnMetaCode/agency-agents-zh

Strengths

  • Covers a wide range of roles, from engineering and security to community and marketing
  • Supports many AI coding tools through one install script, with a dry-run option
  • Each agent file is plain markdown, so it is easy to read, edit, and version

Trade-offs

  • The README describes the agents as prompt-based personalities, not verified tools, so results depend on the AI tool and need human review
  • The README notes a limit of about 119 agents in OpenCode, with the rest silently dropped, so large installs may not fully load there
  • The README does not describe how individual agents were evaluated, so the claims of being production-ready are hard to check from the README alone

Stats synced from GitHub. The assessment above was drafted with AI from the project's README on 2026-10-29 and may miss recent changes — check the README before relying on a command. MIT licence: free to use, modify and distribute, including commercially, as long as you keep the copyright notice.