AgentHub

About AgentHub

AgentHub is the open-source registry for discovering and sharing AI coding agents across multiple platforms — Claude Code, Gemini CLI, and OpenAI Codex CLI. Like Smithery for MCP servers, AgentHub helps developers discover, share, and collect agents for their AI-powered development workflows.

What is AgentHub?

AgentHub is a centralized registry where developers can browse, search, and install AI coding agents. Agents are autonomous AI-powered assistants that can plan, execute, and verify development tasks within CLI-based coding tools. They range from high-level orchestrators to focused specialist workers.

Supported Platforms

Claude Code

Anthropic's CLI coding agent. Agents are defined as .md files in .claude/agents/ directories.

Gemini CLI

Google's CLI coding agent. Agents are defined as .md files in .gemini/agents/ directories.

Codex CLI

OpenAI's CLI coding agent. Agents are defined as .toml files in .codex/agents/ directories.

Agent Categories

Orchestrators

Meta-agents that coordinate multi-agent workflows, classify intent, route to optimal models, and delegate to specialized sub-agents. They manage the big picture.

Specialists

Domain experts with deep knowledge in specific areas like security, architecture, testing, or code review. They provide expert-level analysis and recommendations.

Workers

Focused executors that handle implementation, code generation, refactoring, and other hands-on development tasks. They build and ship.

Analysts

Read-only advisors that provide strategic counsel, code review, performance analysis, and architectural guidance without making changes.

Workflow Stages

Each agent participates in one or more stages of the development workflow. Stages are automatically classified based on the agent's description, tools, and model.

Discover

Codebase exploration, file search, information gathering

Plan

Requirements analysis, architecture design, strategy

Implement

Code writing, feature building, refactoring

Review

Code review, security audit, quality inspection

Verify

Testing, QA, validation, coverage checks

Debug

Debugging, root cause analysis, error tracing

Operate

Git operations, deployment, documentation, releases

Cross-Platform Compatibility

SKILL.md Open Standard

SKILL.md is a shared open standard adopted across Claude Code, Gemini CLI, and Codex CLI. It defines a portable format for agent skills that can work across all three platforms, enabling a unified ecosystem of reusable AI coding capabilities.

Model Tiers

Each platform has its own model ecosystem. Agent definitions specify which model to use within their platform.

Claude Code

Opus

Deep reasoning, complex architecture

Sonnet

Balanced coding, most tasks

Haiku

Fast and lightweight

Gemini CLI

Gemini 2.5 Pro

Deep reasoning and analysis

Gemini 2.5 Flash

Fast and efficient

Codex CLI

GPT-5.4

Strong coding and reasoning

GPT-5.4 Mini

Speed and efficiency

How Auto-Collection Works

1. Discover

The collector scans known GitHub repositories for agent definition files across all supported platforms.

2. Parse

Markdown frontmatter and TOML configs are extracted and normalized into a consistent data model with descriptions, capabilities, and install commands.

3. Merge

New entries are merged with the existing registry, deduplicating by slug. The collection can be triggered manually or via scheduled cron jobs.

How to Contribute

There are several ways to contribute to AgentHub:

  • Submit via the form — Use the Submit page to add your agent directly.
  • Open a Pull Request — Add entries to the JSON data files and submit a PR on GitHub.
  • Report Issues — Found a bug or have a feature request? Open an issue on the GitHub repository.
  • Improve the Platform — The entire application is open source. Contributions to the codebase are welcome.

Tech Stack

Next.js 15TypeScriptTailwind CSS v4shadcn/uiLucide IconsVercel AnalyticsAuth.js v5Vercel

References