Agency Agents: hire a 267-strong AI agency inside your coding agent

Ask a general-purpose assistant to build a financial model, tune a PostgreSQL schema or plan a Baidu SEO campaign, and the answer thins out fast — unless you write the expert persona yourself. agency-agents, by Michael Sitarzewski (MIT, created October 13, 2025; 158,910 stars and 25,656 forks as of October 11, 2026), turns that writing job into a menu: 267 specialist agents organized as a company across 18 divisions, from React frontend and database tuning to brand guardianship and PPC buying. Each agent is one Markdown file with a personality, workflows, concrete deliverables and success metrics. The project grew out of a Reddit thread and a year of iteration.

Core features

• 267 agents in 18 divisions: 65 in engineering, 59 in specialized roles, 37 in marketing — down to 3 in healthcare and 1 in research. The README says 230+; counting the Markdown files in the division directories gives 267 (checked October 11, 2026). The marketing division alone has dedicated agents for Xiaohongshu, Zhihu, Bilibili, Baidu SEO and Kuaishou.

  Division	Agents	Division	Agents
  Engineering	65	Specialized	59
  Marketing	37	GIS	13
  Security	12	Design	10
  Sales	9	Testing	9
  Paid media	7	Project management	7

  The remaining 8 divisions (academic, game development, product, spatial computing, support, finance, healthcare, research) hold 1–6 each, counted from the repo files on October 11, 2026.

• One source, 17 tools: tools.json is the install contract, and convert.sh renders the same roster into each tool’s native format — Claude Code gets per-agent files in ~/.claude/agents/, while Codex, Cursor, Copilot, Gemini CLI, Kimi, DeepSeek Harness and ZCode are all covered.

• Pick your headcount: install whole divisions with --division engineering,security, single agents with --agent frontend-developer,ui-designer, browse with --list teams; strategy/runbooks.json ships prebuilt rosters such as startup-mvp.

• A desktop app if you skip terminals: agencyagents.app browses the full roster and installs into each tool with a click, auto-updates, and ships as a native app for macOS, Linux and Windows (brew install --cask msitarzewski/agency-agents/agency-agents).

Typical use cases

• A one-person company with every department: the official Nexus Spatial Discovery example runs 8 agents at once — product trend researcher, backend architect, brand guardian, growth hacker, UX researcher and more — to evaluate a software opportunity and produce one unified plan.
• Patching a generalist’s gaps: install database-optimizer for slow queries, payments-billing-engineer for Stripe integration, china-network-engineer for Chinese network compliance — one agent at a time.
• Designing multi-agent pipelines: the repo includes a multi-agent systems architect agent whose own specialty is topology, context and failure recovery.

Quick start

git clone https://github.com/msitarzewski/agency-agents.git
cd agency-agents
./scripts/install.sh --tool claude-code

Then say, as the README suggests: activate Frontend Developer mode and help me build a React component. For other tools, run ./scripts/convert.sh once to generate the integration files, then ./scripts/install.sh --tool codex or --tool cursor.

Summary

Worth installing for solo developers and small teams: pick a few agents and your coding agent suddenly has a bench of specialists to summon. Hold off if you expect evaluated, guaranteed expertise — these are carefully written prompt personas, not benchmarks. MIT-licensed, Shell scripts plus Markdown. Three caveats:

• Verify quality yourself: agents are community-contributed and the repo carries 224 open issues; read one or two files before trusting a role. OpenCode has an upstream bug that registers only about 119 agents and silently drops the rest — keep your selection under the limit with --division.
• Don’t install all 267: the full roster bloats your context and tool list; install by division or agent instead.
• The Chinese fork is a separate line: agency-agents-zh (21,176 stars, updated October 8, 2026) adds 64 original agents for the China market and does not sync with upstream.

Same idea, smaller scale: gstack configures Claude Code as an engineering team, and OpenDots is a template for building your own AI employee roster. agency-agents stands out on headcount and its conversion layer for 17 tools.