wshobson/agents

⭐ 35,111  ·  Python  ·  GitHub Repo

Intelligent automation and multi-agent orchestration for Claude Code

agents anthropic anthropic-claude automation claude claude-code claude-code-cli claude-code-commands

1-Sentence Summary

Production-ready multi-agent orchestration ecosystem for Claude Code with 185 specialized agents and 80 plugins.

🔥 Key Capabilities & USP

• Granular Plugin Architecture (80 plugins): Each plugin loads only its specific agents, commands, and skills into Claude Code's context, solving the critical pain point of token bloat. Average 3.6 components per plugin (Anthropic's 2-8 pattern) means you pay only for what you use.
• Multi-Agent Orchestration (185 agents + 16 workflows): Domain-expert agents collaborate through workflow orchestrators for complex operations—full-stack development, security hardening, ML pipelines, and incident response. No more single-agent bottlenecks.
• Agent Skills System (153 modular packages): Progressive disclosure loads specialized expertise only when activated, maintaining token efficiency while providing deep domain knowledge across 25 categories.
• PluginEval Quality Framework: Three-layer evaluation system (static analysis, LLM judge, Monte Carlo simulation) with 10 quality dimensions and anti-pattern detection. Statistical confidence intervals certify plugin/skill quality—not just "it works on my machine."
• Cross-Platform Support: Works as Claude Code plugins and native Gemini CLI extension with 153 on-demand discoverable skills. Future-proof your investment.

USP: The only ecosystem that combines granular plugin isolation, multi-agent orchestration, and a statistically rigorous quality certification framework—all optimized for minimal token consumption in production environments.

Technical Architecture

Component	Details
Language	Python (primary), CLI integrations
Plugin Architecture	Isolated, single-purpose plugins; each contains its own agents, commands, skills
Component Pattern	Anthropic 2-8 component pattern (avg 3.6/plugin)
Knowledge Loading	Progressive disclosure—skills load only on activation
Quality Framework	PluginEval: static analysis → LLM judge → Monte Carlo simulation
Evaluation Dimensions	10 quality dimensions with statistical confidence intervals
Platform Support	Claude Code plugins + Gemini CLI extension
Installation	/plugin marketplace add wshobson/agents + individual plugin installs

Quick Start Guide

# Add the marketplace to Claude Code
/plugin marketplace add wshobson/agents

# Browse available plugins
/plugin

# Install essential development plugins
/plugin install python-development
/plugin install javascript-typescript
/plugin install backend-development
/plugin install kubernetes-operations
/plugin install cloud-infrastructure
/plugin install security-scanning
/plugin install comprehensive-review
/plugin install full-stack-orchestration

# Install advanced orchestration plugins
/plugin install agent-teams@claude-code-workflows
/plugin install conductor@claude-code-workflows
/plugin install plugin-eval@claude-code-workflows

# Gemini CLI alternative
gemini extensions install https://github.com/wshobson/agents

# Troubleshooting - clear cache if needed
rm -rf ~/.claude/plugins/cache/claude-code-workflows && rm ~/.claude/plugins/installed_plugins.json

# Evaluate plugin/skill quality
uv run plugin-eval score path/to/skill --depth quick
uv run plugin-eval score path/to/skill --depth standard
uv run plugin-eval certify path/to/skill

Pros, Cons & Use Cases

Pros

• Token efficiency: Isolated plugin loading means minimal context overhead—critical for Claude Code's token limits
• Comprehensive coverage: 25 categories spanning architecture, languages, infrastructure, quality, data/AI, documentation, business ops, and SEO
• Production-ready quality: PluginEval framework with statistical rigor ensures certified plugins meet quality standards
• Cross-platform flexibility: Works with both Claude Code and Gemini CLI—no vendor lock-in
• Progressive disclosure: Skills activate only when needed, keeping context lean

Cons

• Environment dependency: Requires Claude Code or Gemini CLI—not a standalone tool
• Installation friction: Must use plugin names (not agent names) for installation; naming mismatch can cause confusion
• Cache issues: May require manual cache clearing (rm -rf ~/.claude/plugins/cache/...) for troubleshooting
• Learning curve: 185 agents, 80 plugins, 153 skills—new users may feel overwhelmed without guided onboarding

Who should NOT use this?

• Developers not using Claude Code or Gemini CLI: This ecosystem is tightly coupled to these environments; standalone use is not supported
• Teams seeking a single "all-in-one" plugin: The granular architecture intentionally avoids monolithic plugins; if you want one install for everything, this isn't it
• Users with minimal token budget concerns: If you're not optimizing for token consumption, the progressive disclosure and isolation benefits add unnecessary complexity
• Prototyping/toy projects: The 185-agent ecosystem is designed for production workflows; simpler projects may find it over-engineered

Ideal Use Cases

• Full-stack development teams needing specialized agents for frontend, backend, infrastructure, and security—all collaborating through orchestrators
• Security-conscious organizations requiring hardened, certified plugins with statistical quality guarantees (PluginEval framework)
• Multi-platform teams running both Claude Code and Gemini CLI who want a unified agent ecosystem
• Complex workflow automation: Incident response pipelines, ML model deployment, multi-service architecture reviews—where single-agent approaches fail
• Token-optimized environments: Teams pushing Claude Code's context limits who need maximum capability per token

Community & Activity

With 35,111 stars and active maintenance through May 2026, this project has clearly resonated with the developer community. The ecosystem's rapid growth—from initial release to 80 plugins, 185 agents, and 153 skills—demonstrates strong contributor momentum and real-world adoption. The inclusion of a formal quality evaluation framework (PluginEval) suggests the maintainers are committed to production-grade reliability, not just feature quantity. This isn't a side project; it's becoming the de facto standard for Claude Code multi-agent orchestration.