msitarzewski/agency-agents

A complete AI agency at your fingertips - From frontend wizards to Reddit community ninjas, from whimsy injectors to reality checkers. Each agent is a specialized expert with personality, processes, and proven deliverables

Summary

agency-agents is a collection of specialized AI agent personas stored as Markdown files, organized into divisions such as Engineering, Design, Paid Media, Sales, and Marketing. Each agent profile bundles an identity, mission, workflows, technical examples, and success metrics, and the repo ships with installer scripts that drop the files into many AI coding tools. The project is MIT-licensed and written primarily in Shell (the install/convert scripts).

What it is useful for

The repo is useful when you want pre-built, role-specific AI agents instead of writing your own prompts from scratch. The Engineering division alone covers roles like Frontend Developer, Backend Architect, SRE, Database Optimizer, Code Reviewer, Incident Response Commander, Payments & Billing Engineer, and Search Relevance Engineer, so a team can pick the specialists that match a current project. Other divisions extend the same idea outside pure coding, with agents for UX research, paid-media auditing, and sales coaching.

Because each agent is a standalone Markdown file, the same persona can travel across the tools your team already uses. The README lists supported targets including Claude Code, Cursor, Codex, Gemini CLI, OpenCode, Qwen, Osaurus, GitHub Copilot, Antigravity, Aider, Windsurf, Kimi Code, Hermes, Mistral Vibe, and DeepSeek Harness. You can also treat the files as reference material, copying or adapting just the roles you need.

How engineers can use it

The README documents two entry points: a desktop app (also available via brew install --cask msitarzewski/agency-agents/agency-agents) and the bundled scripts. For a script-based install into Claude Code:

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

You can narrow the selection to a single division or specific agents, list what is available, and dry-run before writing files:

./scripts/install.sh --list teams
./scripts/install.sh --tool claude-code --division engineering,security
./scripts/install.sh --tool opencode --division engineering --dry-run

For other tools, the README describes a two-step flow: run ./scripts/convert.sh to generate integration files, then ./scripts/install.sh (with a target flag such as --tool antigravity, --tool gemini-cli, --tool cursor, or --tool copilot). If you only want one division, the README also documents a manual copy, e.g. cp engineering/*.md ~/.claude/agents/, after which you can activate an agent in a session by name.

A documented limitation: OpenCode's runtime currently registers only about 119 agents and silently drops the rest, so the README recommends using --division to keep installations under that cap, and the installer warns when a selection would exceed it. For the full matrix of supported tools and the runbooks.json workflow for installing pre-rostered teams, see the README's "Multi-Tool Integrations" section.