Your AI assistant’s org chart, written in Markdown

What it does

The Agency is a collection of markdown files, each defining a specialist persona—Frontend Developer, UX Researcher, PPC Strategist, Incident Response Commander, and roughly thirty others. Shell scripts convert and drop these files into directories that Claude Code, Cursor, GitHub Copilot, Aider, Windsurf, and a handful of other tools already read for system prompts. It is, in effect, a staffing agency where the employees are static text files.

The interesting bit

Instead of treating prompt engineering like a dark art of magic words, the project frames it as organizational design. Every agent carries an identity, core mission, expected deliverables, and success metrics. The shell glue auto-detects which AI tools you have installed and copies the personas to the right config paths, which saves you from maintaining a personal spreadsheet of copy-pasted prompts.

Key highlights

• ~30+ roles across Engineering, Design, Sales, and Paid Media, including niche specialists like Feishu Integration Developer, Whimsy Injector, and Solidity Smart Contract Engineer
• Each agent file bundles personality traits, technical workflows, code examples, and communication style
• install.sh and convert.sh support Claude Code, Cursor, Copilot, Aider, Windsurf, Gemini CLI, and several others
• MIT licensed and openly iterated; the README notes it originated from a Reddit thread
• Deliverable-focused rather than open-ended: agents are scoped to measurable outcomes

Caveats

• The repository is fundamentally a curated prompt library wrapped in shell scripts; the “production-ready” and “battle-tested” claims are the author’s own and not independently substantiated in the sources
• The README catalogs titles and one-line specialties but does not expose the actual prompt text, so depth and quality of any single agent is impossible to verify without cloning
• There is no visible evaluation framework or automated testing to confirm the personas improve output over baseline prompts

Verdict

Worth a look if you spend your day in Claude Code or Cursor and want pre-cooked specialist personas without drafting system prompts from scratch. Give it a pass if you are hunting for autonomous runtime agents, orchestration logic, or anything more dynamic than well-structured markdown.