wshobson/agents

What it solves

It provides a standardized, portable library of agentic building blocks—including specialized agents, skills, and commands—that can be deployed across multiple different AI coding assistants and CLI tools without needing to rewrite them for each platform.

How it works

The project uses a single Markdown-based source of truth for all plugins. It then employs a set of adapters (harnesses) to generate native artifacts for various platforms, such as Claude Code, Cursor, Codex CLI, and GitHub Copilot. This ensures that each tool receives the configuration in its own idiomatic format rather than a generic translation.

Plugins are isolated and composable, meaning that when a user installs a specific plugin, only the relevant components (agents, commands, and skills) are loaded into the AI's context to avoid bloat.

Who it’s for

Developers and power users of AI coding agents who want to extend their tools with domain-specific expertise (e.g., security, infrastructure, or specific programming languages) across different AI environments.

Highlights

• Massive Library: Includes 94 plugins, 202 agents, 183 skills, and 105 commands.
• Multi-Harness Support: Natively supports Claude Code, Codex CLI, Cursor, OpenCode, Antigravity CLI, GitHub Copilot, and Pi.
• Tiered Model Strategy: Maps tasks to different model tiers (from Fable 5 for long-horizon work to Haiku for fast operational tasks).
• Quality Evaluation: Features a three-layer framework (plugin-eval) using static analysis, LLM judges, and Monte Carlo simulations to certify plugin quality.