Open-source collection of production-ready subagents for Claude Code. Drop a .md file into your project — get a specialized AI teammate.
Subagents are specialized AI assistants that run inside Claude Code with their own isolated context window, custom system prompt, and scoped tool access. Instead of explaining your stack every session, you give Claude a permanent expert it can delegate to.
You ──→ Claude Code ──→ 🏗️ nest-architect (designs modules, enforces patterns)
──→ 🔒 security-reviewer (audits code for vulnerabilities)
──→ 🧪 test-writer (generates test suites)
Install a single agent:
# Project-level (only this repo)
mkdir -p .claude/agents
cp agents/nest-architect.md .claude/agents/
# Global (all your projects)
mkdir -p ~/.claude/agents
cp agents/nest-architect.md ~/.claude/agents/Use it:
> /agents # see all available agents
> Use nest-architect to add a Product module
> Have the nest-architect review my PR
Claude will also auto-delegate when a task matches the agent's description.
| Agent | Stack | Description |
|---|---|---|
| nest-architect | NestJS, Prisma, Pino | Clean Architecture enforcer. Knows layers, patterns, entity scaffolding. |
Every agent is a single Markdown file with YAML frontmatter:
---
name: my-agent # unique identifier
description: > # WHEN Claude should use this agent
Use when designing new modules,
reviewing architecture, ...
tools: Read, Glob, Grep # tool access (scope it down!)
model: sonnet # sonnet (fast) or opus (powerful)
---
You are a senior [role]... # system prompt starts here
## Project Architecture
...
## Your Operating Rules
...- One goal per agent. "Security reviewer", not "universal helper".
- Scope the tools. Read-only agents get
Read, Glob, Grep. Builders getWrite, Edit, Bash. - Be opinionated. Agents work best when they enforce specific patterns, not generic advice.
- Include examples. Show the exact code patterns, file structures, and naming conventions.
- State weaknesses. Add
"Be critical, don't be agreeable"— LLMs default to politeness.
Option 1: Let Claude generate it
> /agents → Create New Agent → Generate with Claude
> "Create a security review agent that checks for OWASP Top 10 vulnerabilities"
Option 2: Copy and adapt
- Pick an agent from this repo that's closest to your needs
- Copy it to
.claude/agents/or~/.claude/agents/ - Edit the system prompt to match your project's stack and conventions
Option 3: Build from your codebase
Ask Claude Code to analyze your project and generate an agent:
> Analyze this codebase architecture and create a subagent that enforces
> our patterns. Save it to .claude/agents/my-architect.md
awesome-claude-code-agents/
├── agents/ # ready-to-use agent files
│ └── nest-architect.md
├── templates/ # starter templates
│ └── basic-agent.md
├── CONTRIBUTING.md
└── README.md
We welcome contributions! To add a new agent:
- Fork this repo
- Create your agent in
agents/your-agent-name.md - Follow the agent anatomy format
- Include a real-world description of what stack/project it targets
- Open a Pull Request with:
- What the agent does
- What stack it's built for
- Example usage prompts
- Agent has a clear, specific
descriptionfield - Tools are scoped to minimum required access
- System prompt includes concrete code patterns (not just generic advice)
- Tested in at least one real project
- File is under 500 lines (Anthropic recommendation)
- Claude Code Subagents Docs — official documentation
- Claude Code Agent Teams — multi-agent orchestration
- Claude Code Best Practices — Anthropic's engineering guide
MIT — use these agents however you want, in personal or commercial projects.
Built with Claude Code ⚡ Contributions are welcome — let's build the best agent library together.