Agency Agents: 166k stars, 230+ specialist agents —
the catch?
Agency Agents is an open-source roster of 230+ specialist AI agent personalities you install into Claude Code, Cursor, Codex and more — the catch: stars are reach, not proof, and each agent runs as you, on your own bill.
- Claim
- 166 THOUSAND STARS 230+ AGENTS?
- Catch
- NO BENCHMARKS
- Verdict
- WORTH BROWSING, your rules
One hundred and sixty-six thousand five hundred forty-six stars. That is what GitHub's API shows for msitarzewski/agency-agents as of 5 November 2026 — a roster of AI agent personalities that pitches itself as "a complete AI agency at your fingertips," with 230+ specialized agents across engineering, design, sales, marketing, research, healthcare and more.
The pitch, in the README's words: browse the roster, pick the specialists you need (Frontend Developer, Backend Architect, Security Engineer, Research Synthesist and the rest), and install them into your own coding agent — Claude Code, Cursor, Codex, Gemini CLI, OpenCode, Windsurf and others. Each agent is a Markdown definition file plus Shell install scripts (./scripts/install.sh --tool <name>), so you bring the agent and your own signed-in account; the repo supplies the personality, not the software.
What it actually is
It is not autonomous software, and it is not one product you download. It is a collection of agent definitions — roughly ten thousand lines of personality, process and code examples across more than a dozen folders — plus a set of conversion and install scripts that turn those Markdown files into the format each coding tool expects (.claude/agents/, .cursor/rules/, Codex TOML, Qwen SubAgents, and more). MIT license, created 13 October 2025, last pushed 25 October 2026, 26,974 forks.
The "install with a click" experience is something else entirely: a separate native desktop app called Agency Agents (agencyagents.app, a different repository) for macOS, Linux and Windows that browses the roster and installs agents with auto-updates. The repo itself is the command-line route.
What you need
Your own signed-in coding agent — Claude Code is supported first — and your own LLM bill. Each agent acts through your account, under your own agreement with the provider, and you pay for every token it uses. The skill itself is free and MIT-licensed; one-click install only exists in the separate app.
The catch
Verdict
Agency Agents is a genuine, well-organized roster and the engineering is open for anyone to read — but stars are reach, not proof. There are no benchmarks showing these agents ship better code, each one acts as you with your permissions, and the one-click install lives in a separate app. Worth browsing if you already use Claude Code or Cursor; not a shortcut to an AI team.
Sources: github.com/msitarzewski/agency-agents (README, MIT, checked 2026-11-05; 166,546 stars, 26,974 forks, created 2025-10-13, Shell) · github.com/msitarzewski/agency-agents-app · agencyagents.app · GitHub API, 5 November 2026
SOURCES
THE CATCH BY EMAIL
Get the catch by email
One short email a week: the viral claims we checked, what the sources really say, and the catch. No spam, unsubscribe in one click.
Free. Sent through Buttondown. Privacy
