Agency Agents: 144 AI Role Templates with One-Click Installation for Cursor/Claude Code

Every time you ask AI to help write code, create designs, or review tests, do you find yourself starting with "You are a senior engineer"? Then spending tons of time filling in details — what to ask first, what to deliver, what counts as done? The Agency Agents open-source project has completely standardized this repetitive work.

Not a Chatbot — A Structured AI Role Asset Library

Many people still use AI at the "one-sentence persona" stage: tell the AI it's a certain role, then hope it gives professional answers. This approach works but is too crude. What truly determines AI output quality is an entire set of detailed definitions behind the role — what's the core mission, how does the workflow go, what are the deliverables, how are success metrics measured, and what's the communication style.

This reflects the broader trend of Prompt Engineering evolving toward engineering-grade management. Traditional AI prompts are typically improvised natural language instructions lacking standardized structure. As large language models penetrate enterprise scenarios more deeply, the industry has gradually recognized that prompts themselves need to be modularly managed like code — decomposing role definitions, task constraints, output formats, and evaluation criteria into independent configurable modules to achieve reusability and maintainability. This closely parallels the evolution in software engineering from scripted programming to modular architecture.

What Agency Agents does is organize these details into structured Markdown files, forming an installable, copyable, and customizable open-source AI role directory. The repository currently contains 144 specialized Agent role templates spanning 12 business domains, including engineering, design, marketing, product, project management, testing, support, finance, game development, and academia.

What's Actually Inside the 144 AI Agent Role Templates

This isn't a simple prompt collection. Each role file contains comprehensive responsibility definitions, far richer than "you are XX expert."

In multi-Agent systems research, a high-quality Agent definition typically includes several key dimensions: the core Mission defines the Agent's fundamental purpose; the Workflow specifies task execution steps and decision logic; Deliverables clarify the specific forms and format requirements of outputs; Success Criteria provide quantitative or qualitative standards for evaluating output quality; and Communication Style constrains the Agent's language expression and interaction patterns. Together, these dimensions form the Agent's "behavioral contract," transforming AI output from high-randomness free-form responses into predictable, auditable, standardized production.

Engineering: Covering the Full Software Development Chain

The engineering directory covers key roles across the entire software development lifecycle: Front-end Developer, Back-end Architect, Security Engineer, Code Reviewer, and more. Each role defines not just identity, but also work processes, delivery standards, and risk boundaries.

Modern software development has evolved from "one full-stack person handles everything" to a highly specialized division of labor. Front-end developers handle user interface interactions, back-end architects design server-side structures, data flows, and API interfaces, security engineers focus on vulnerability detection, threat modeling, and compliance audits, and code reviewers gate code quality, maintainability, and team conventions. In real teams, these roles have clear responsibility boundaries and collaboration interfaces. Agency Agents maps this mature division-of-labor model onto AI role definitions, enabling each AI Agent to focus on a specific professional domain rather than trying to be a "knows everything but masters nothing" generalist.

Design: Clear Responsibility Division

The design directory includes UI Designer, UX Researcher, Brand Guardian, and other roles. These roles have clear responsibility divisions, eliminating the gray areas under the vague umbrella concept of "design."

Testing & Quality: Far More Than "Run It Once and See If There Are Bugs"

The testing directory's role design shows considerable depth, including Evidence Collector, Reality Checker, API Tester, and more. These names themselves reflect deep understanding of testing work — testing is far more than "run it once and see if there are bugs."

The software testing field is far more complex than most people realize. Modern testing theory divides testing activities into multiple layers: unit tests verify the correctness of the smallest code units, integration tests check inter-module collaboration, end-to-end tests simulate real user scenarios, and API tests verify interface contract consistency. Names like "Evidence Collector" and "Reality Checker" reflect higher-level thinking in test engineering — testing isn't just about finding bugs, it's about systematically collecting evidence to prove or disprove that software behavior meets expectations. This Evidence-Based Testing methodology emphasizes traceability and reproducibility, highly aligned with the structured philosophy of role definitions in Agency Agents.

Product Prototype Development in Practice: Multi-Role Collaboration Workflow Demo

The best way to understand Agency Agents is to treat it as your AI work team asset library. Suppose you're building a new product prototype — the workflow can be organized like this:

1. First, have the Rapid Prototyper help you quickly build a runnable version
2. Then have the Front-end Developer handle the interface implementation
3. Let the Back-end Architect review API and data structure design
4. Have the UI Designer unify the visual experience
5. Finally, let the Code Reviewer do pre-launch quality checks

When switching roles at each stage, you don't need to write prompts from scratch — just load the corresponding Agent file. These files already define what that role should focus on, what to ignore, and what to deliver.

The value isn't in automatically completing all tasks, but in not having to design these roles' responsibility boundaries from zero. Starting with ready-made role templates and then editing them according to your team's habits is far more efficient than reinventing the wheel every time.

Multi-Platform Compatible: Not Locked to Any AI Coding Tool

Agency Agents takes a pragmatic approach to tool compatibility, supporting current mainstream AI coding tools with straightforward installation methods:

• Claude Code: Run the install.sh script provided in the repository, or manually copy Markdown files to your local Agents directory
• Cursor, Windsurf, Aider: The repository provides Convert and Install scripts that automatically transform role files into formats readable by each tool
• Gemini CLI, Quant Code, Kimi Code, etc.: Format conversion and installation are also supported

The AI coding tool market is currently in a phase of rapid differentiation and competition. Claude Code is Anthropic's command-line AI coding assistant that collaborates with developers directly in the terminal; Cursor is a VS Code-based AI-enhanced editor that deeply integrates LLMs into the code editing workflow; Windsurf (launched by Codeium) is similarly an AI-native IDE; Aider is an open-source command-line AI pair programming tool supporting multiple model backends; Gemini CLI is Google's command-line AI tool. These tools each have different configuration formats and Agent loading mechanisms, which is why Agency Agents needs to provide format conversion scripts — it solves the portability problem of role assets across heterogeneous tool ecosystems.

The same set of role assets can migrate between different tools. When you switch from one AI coding tool to another, your accumulated role library won't go to waste.

Understand the Boundaries Before Using Agency Agents

After covering the strengths, let's be clear about what Agency Agents is not:

• Not a SaaS product: No online platform, no paid subscription
• Not a workflow orchestration engine: It won't automatically arrange the collaboration sequence of multiple Agents
• No guarantee of result quality: The AI model you're using is still the one actually executing tasks

It's more like a set of structured, auditable, version-manageable role prompt templates and process specifications. You should treat it as a starting point rather than an endpoint, continuously iterating these Agent files according to your project needs, quality standards, and team language.

Who Benefits Most from This AI Role Template Library

Based on the project's characteristics, three types of users can extract the most value from Agency Agents:

Heavy AI users. People who frequently use AI to write code, create designs, or draft documents, but are tired of repeatedly writing role settings every time. Agency Agents lets you standardize prompt templates — define once, reuse repeatedly.

Small teams or indie developers. Those who want to clearly separate product, engineering, testing, and marketing roles but don't yet have a complete team. Through the AI role library, one person can simulate a multi-role collaborative working mode.

Multi-Agent workflow researchers. Those who want to see how a larger-scale role library organizes categories, defines deliverables, and integrates tools. The organizational approach of 144 roles across 12 domains is itself an excellent reference architecture.

If you only occasionally ask AI a small question, this project might feel like overkill. But if you've already started treating AI as a long-term collaboration partner, it's very much worth exploring in depth.

Core Value: From One-Sentence Personas to Manageable AI Role Assets

The core value of Agency Agents isn't simply having 100+ more prompt templates. The real problem it solves is: crystallizing professional roles' responsibilities, processes, and quality standards into files that can be installed, copied, modified, and version-managed.

This shares the same philosophy as "Infrastructure as Code" (IaC) in software engineering. IaC is one of the core concepts in DevOps, advocating that infrastructure definitions like server configurations, network topologies, and deployment processes should be described in code files rather than managed through manual operations. Tools like Terraform, Ansible, and Pulumi are typical implementations of this philosophy. The core benefits of IaC include: version traceability (managing change history through Git), environment reproducibility (anyone can rebuild the same environment from the same configuration file), and change auditability (every modification has records and approval processes). When your AI role settings also become manageable code assets like this, team collaboration consistency, maintainability, and scalability all improve significantly.

The next time you want AI to play the role of front-end, back-end, design, testing, or growth — you don't have to start from "you are an expert" every time. Grab a ready-made AI work team first, then tune it to fit your needs — that's the right way to use Agency Agents.