TAU-HOME.COM
LOADING

wshobson/agents: Unified Agent and Skill Catalog for Claude Code, Cursor, and Codex

An open-source multi-harness plugin and skill marketplace supporting Claude Code, Codex CLI, Cursor, Antigravity, GitHub Copilot, and Pi from a single Markdown

tau · November 12, 2026

#claude-code #cursor #coding-agents #developer-tools #ai-skills

wshobson/agents: Unified Agent and Skill Catalog for Claude Code, Cursor, and Codex

The open-source project wshobson/agents has gained traction on GitHub by providing a unified catalog of AI coding agents, skills, and commands written once in Markdown and compiled for Claude Code, OpenAI Codex CLI, Cursor, Antigravity, GitHub Copilot, and Pi.

Original editorial illustration: AGENT CATALOG

Hero image: Original TAU HOME editorial concept illustration, not a product screenshot. The repository addresses harness fragmentation across AI programming tools, where developers previously had to rewrite instructions, tool permissions, and skill structures for each environment.

Single Markdown Source and Multi-Harness Adapters

wshobson/agents maintains its primary components in the Claude Code plugin Markdown specification. Adapters located in tools/adapters/ parse these sources and output the artifact formats required by each supported coding harness.

  • Catalog Scope: Contains 94 plugins (92 local plugins and 2 external entries), covering 202 subagents, 185 skills, and 105 workflow commands.
  • Model Alias Translation: Source agents declare model aliases such as opus, sonnet, haiku, fable, or inherit. Adapters translate these aliases into valid model configurations supported by each target harness.
  • Permission Mapping: Adapters convert tool permission configurations where the target tool allows it, while repository CI validates generated artifacts and tests CLI compatibility.

Output behavior varies across platforms. Cursor consumes an committed marketplace registry, while Codex CLI provides native manifests alongside generated TOML agents, automatically splitting oversized skill bodies into reference files to comply with its 8 KB prompt budget.

Installation Routes and Setup

The project supports three distinct installation routes based on your target tool.

1. Native Installation on Claude Code and Codex CLI

Within Claude Code, register the marketplace and install plugins directly:

/plugin marketplace add wshobson/agents
/plugin install python-development@claude-code-workflows

Once installed, invoke scaffolding commands or assign tasks to bundled specialist agents:

/python-development:python-scaffold Create a FastAPI service with tests

For OpenAI Codex CLI, run the native marketplace commands in your terminal:

codex plugin marketplace add wshobson/agents
codex plugin add python-development@claude-code-workflows

2. Standalone Skill Installation

To use specific skills without installing full plugin bundles, commands, or hooks, use either GitHub CLI or the npx skills installer directly against the remote repository:

gh skill install wshobson/agents python-testing-patterns
npx skills add wshobson/agents --skill python-testing-patterns

The gh skill route requires GitHub CLI version 2.90 or later, while npx skills requires a Node.js and npm runtime.

3. Local Clone Installation

For Antigravity, Antigravity CLI, GitHub Copilot, and Pi, clone the repository and run the Makefile installer for your environment. This workflow requires Python 3.12 or later and Astral uv.

gh repo clone wshobson/agents ~/agents
cd ~/agents
make install-copilot
make install-antigravity
make install-pi

Known Limitations and Caveats

Keep the following operational constraints in mind before deploying these plugins into production pipelines:

  • Pi Subagent Extension Requirement: Running agents in Pi requires installing its reference subagent extension or an equivalent compatible extension.
  • Experimental Evaluation Layer: The embedded plugin-eval package includes static checks, but its LLM judge and Monte Carlo evaluation layers are experimental and not validated against human ground-truth labels. Running LLM evaluations incurs API billing through ANTHROPIC_API_KEY or an active Claude Code login session.
  • Pensyve Cloud Sunsetting: The external memory plugin Pensyve closed its managed cloud service on October 1, 2026. Hosted dashboards, APIs, and cloud MCP endpoints are offline; integrating Pensyve now requires self-hosting the open-source runtime.

The repository is distributed under the MIT License, while external plugins retain their respective source licenses.

Sources

Impact-Site-Verification: 9b3ebd10-6c45-44cf-a014-5b8c19b7b285