Claude Code agents are autonomous AI workers that can plan, execute, and iterate on complex multi-step development tasks without constant human supervision. Agents go beyond single prompts by breaking down high-level goals into sub-tasks, executing shell commands, editing files, running tests, and self-correcting errors across an entire development workflow.
AGT40 custom agents
27 Claude Code Agents — AI-Powered Development Teams
Agents represent the most advanced way to use Claude Code. While a prompt gives Claude a single instruction and a skill teaches Claude a repeatable workflow, an agent is a fully autonomous worker that can plan, execute, verify, and iterate on complex multi-step tasks without constant human supervision. Agents turn Claude Code from a conversational assistant into an AI-powered team member.
The architecture behind Claude Code agents is deceptively simple. You provide the agent with a goal, a set of available tools, and constraints. The agent breaks the goal into sub-tasks, determines the optimal execution order, and works through each step using the same tools a human developer would: reading files, running shell commands, editing code, executing tests, and searching documentation. When an agent encounters an error, it diagnoses the problem and adapts its approach rather than stopping.
One of the most powerful patterns in the Claude Code ecosystem is multi-agent orchestration. A lead agent can dispatch multiple sub-agents to work on independent tasks simultaneously. Picture a scenario where you ask a single agent to prepare a release: it spawns one sub-agent to run the full test suite, another to generate the changelog from commit history, a third to update version numbers across the codebase, and a fourth to check for dependency vulnerabilities. Each sub-agent works in its own context, and the lead agent merges the results and handles conflicts.
This directory features 27 battle-tested agents covering the full development lifecycle. You will find code review agents that catch bugs, security holes, and architectural anti-patterns, testing agents that generate comprehensive test suites and maintain coverage thresholds, refactoring agents that modernize legacy code while preserving behavior, and DevOps agents that manage deployments, infrastructure, and monitoring.
The best agents in this collection come with well-defined boundaries. They specify exactly what they will and will not do, include safety checks before destructive operations, and produce clear logs of every action taken. Whether you are a solo developer looking to multiply your output or a team lead wanting to automate repetitive workflows, these agents give you a concrete starting point to build your own AI-powered development team.
Quick Comparison: Top 5 Claude Code Agents
Name
Specialty
Best For
Difficulty
Code Review Agent
Security and quality analysis
Automated PR reviews on every push
Intermediate
Test Writer Agent
Test generation and coverage
Maintaining 80%+ test coverage automatically
Intermediate
Refactor Agent
Legacy code modernization
Migrating codebases to modern patterns safely
Advanced
DevOps Agent
Deployment and infrastructure
Automated releases with rollback safety
Advanced
Documentation Agent
API docs and changelogs
Keeping docs in sync with code changes
Beginner
How to Create Custom Agents
Building a custom agent starts with defining its purpose, available tools, and success criteria in a CLAUDE.md configuration. You can launch agents via the CLI in non-interactive mode for CI/CD integration, or programmatically through the Claude Code SDK.
# Run an agent in non-interactive (headless) mode
claude --print "Review all files in src/ for security vulnerabilities. \
Report findings in a markdown table with severity ratings."
# Run an agent with a specific CLAUDE.md configuration
claude --print --config .claude/agents/code-review.md \
"Review the changes in the last 3 commits"
# Spawn a sub-agent for parallel task execution
claude --print "Run tests in parallel: unit tests, integration tests, \
and lint checks. Report all results together."
For production agents, define clear boundaries in your CLAUDE.md: specify what the agent should and should not modify, include safety checks before destructive operations, and require confirmation for actions that affect production systems.
Agents vs Skills: When to Use Each
AAgents
Autonomous, multi-step task execution
Self-correcting when errors occur
Can spawn sub-agents for parallel work
Best for complex, open-ended goals
Run in headless or interactive mode
SSkills
Instruction-based, repeatable workflows
Follow predefined steps precisely
Invoked via slash commands
Best for standardized, well-defined tasks
No code required to create
Use skills when the task is well-defined and repeatable, like generating commit messages or running a code review checklist. Use agents when the task is open-ended and requires planning, iteration, and judgment, like migrating a codebase from JavaScript to TypeScript or preparing a full release.
All Agents
40 resources
AutoGPT
Advanced
Autonomous AI agent platform for building and deploying continuous AI agents.
Hermes Agent — Self-Improving AI Agent by Nous Research
Intermediate
A self-improving autonomous AI agent with a built-in learning loop that creates and refines its own skills from experience. Features persistent cross-session memory, multi-platform messaging integration (Telegram, Discord, Slack, WhatsApp, Signal), flexible LLM provider support (Nous Portal, OpenRouter, OpenAI, Anthropic), built-in cron scheduler for unattended tasks, FTS5 session search, and multiple deployment options including Docker, SSH, VPS, GPU clusters, and serverless (Daytona, Modal). 25,900+ stars and 258 contributors. MIT licensed.
40+ specialized marketing skills for AI coding agents covering CRO, copywriting, SEO, analytics, and growth engineering. Built for technical marketers and founders. Works with Claude Code, Codex, and Cursor.
Transforms Claude Code into a full game development studio with 48 specialized agents in a studio hierarchy, 37 slash commands, 8 hooks, and 29 templates. Supports Godot 4, Unity, and Unreal Engine 5.
Bridge connecting local AI coding agents to 10+ messaging platforms (Slack, Telegram, Discord, LINE, WeChat Work, Feishu). Chat with your AI dev assistant from anywhere without a public IP.
Claude Code skills that build complete Godot 4 projects from a game description. AI pipeline handles architecture, art generation, code writing, screenshot capture, and debugging in a single context window.
Specialized agent configuration for backend development. Expert in database design, API architecture, microservices patterns, caching strategies, and security hardening.
Agent tuned for frontend development. Deep knowledge of React, Vue, Svelte, CSS architecture, accessibility standards, performance optimization, and design systems.
Control Claude Code remotely via email, Discord, and Telegram. Start tasks locally, receive notifications when Claude completes them, and send new commands by simply replying to messages from anywhere.
Agent specialized in infrastructure and DevOps. Handles Terraform, Kubernetes, CI/CD pipelines, monitoring setup, and incident response with security-first approach.
Agent configured for data engineering tasks. Specializes in ETL pipelines, data modeling, SQL optimization, dbt workflows, and data quality frameworks.
Claude Code workspace that transforms ideas into multi-format content: research to long-form articles to platform-specific versions (LinkedIn, newsletter, social media, podcast Q&A). Learns your voice.
AI media toolkit skill for Claude Code providing image generation (Flux), text-to-speech, YouTube/audio transcription, OCR, video generation, upscaling, and background removal via deAPI.
Monitor and chat with AI coding agents from Telegram, Discord, and Feishu. Web-based terminal interface with real-time tool visibility and permission approval via mobile devices.
Monitor and interact with Claude Code sessions from Slack, Discord, or Telegram. Respond from your phone while AFK, enabling remote control of local coding sessions through popular chat platforms.
100 production-ready AI skills across 15 categories including sales intelligence, marketing automation, sports analytics, and creative tools. Features meta-skills for conversation archaeology and skill navigation.
A Claude Code agent is an autonomous or semi-autonomous AI worker built on top of Claude Code. Unlike a single prompt interaction, an agent can plan multi-step tasks, execute shell commands, read and write files, run tests, and iterate on its own output. Agents are typically configured through CLAUDE.md files and launched via the Claude Code CLI with specific instructions for a task like code review, refactoring, or deployment.
How do agents differ from prompts?
Prompts are single-turn instructions that shape how Claude responds. Agents are persistent, task-oriented workers that can take multiple actions over time. An agent might receive a high-level goal ("fix all failing tests in the auth module"), then autonomously plan the steps, execute them, verify the results, and report back. Agents use tools like Bash, file editing, and web search throughout their execution.
Can agents work together?
Yes. Claude Code supports multi-agent orchestration where a primary agent dispatches sub-agents to handle independent tasks in parallel. For example, a lead agent might spawn one sub-agent to fix lint errors, another to write missing tests, and a third to update documentation. Each sub-agent works in its own context, and the lead agent coordinates results. This pattern dramatically speeds up large tasks.
How do I build a custom agent?
Start by defining the agent's purpose and scope in a CLAUDE.md or configuration file. Specify what tools the agent should use, what constraints it must follow, and what success looks like. You can launch agents via the CLI using "claude --print" for non-interactive mode or programmatically through the Claude Code SDK. For complex agents, use the skills system to break down the agent's capabilities into modular, testable units.
What are the best agents for code review?
The most popular code review agents in this directory combine static analysis awareness with deep code understanding. Top-rated options include agents that check for security vulnerabilities, enforce architectural patterns, verify test coverage, and even suggest performance optimizations. Many teams configure review agents to run automatically on pull requests via CI/CD integration.
Know an agent we're missing?
Help us build the most complete directory of Claude Code agents.
