wshobson / agents

Plugins, agents, and skills for AI coding tools

Use one collection of plugins, agents, skills, and commands across Claude Code, OpenAI Codex CLI, Cursor, OpenCode, Antigravity CLI, GitHub Copilot, and Pi. A harness is the coding tool that loads and runs the components.

Contributors write each component once, in Markdown, in the Claude Code plugin format. Adapters under tools/adapters/ generate the format each harness reads, map model aliases to that harness’s models, and translate tool permissions where the harness supports them. CI regenerates the artifacts for every harness, validates them, and runs the OpenCode, Antigravity CLI, and Pi artifacts through their installed CLIs. The capability matrix (opens in a new tab) lists what each harness supports, and ARCHITECTURE.md (opens in a new tab) describes the design.

The catalog contains 94 plugins, including 92 local plugins and 2 external entries. The local source contains 202 agents, 185 skills, and 105 commands. Plugins cover Python and JavaScript development, code review, testing, infrastructure, security, and other work.

Multi-harness support

Choose the harness you use and follow its setup route. Available components depend on the installation route and the harness’s capabilities.

See the capability matrix (opens in a new tab) for model mappings, tool permissions, hooks, and installation differences.

For individual skills, use a skills-only installer and select your coding tool.

How it works

A plugin groups related agents, skills, and commands under plugins/. For example, python-development contains three agents, sixteen skills, and a scaffolding command. The chosen harness loads the installed components according to its own discovery and invocation rules.

Contributors maintain one Markdown source in the Claude Code plugin format. Native registries point to that source where supported, and adapters generate the formats needed by each harness. Commands may become skills or prompt templates, and agent formats, model choices, and tool permissions vary by harness.

Source agents specify model aliases such as opus, sonnet, haiku, fable, or inherit. Adapters map those aliases to each harness’s model settings, including fixed defaults for some inherit entries. See model configuration (opens in a new tab) and the adapter mappings (opens in a new tab).

Quality evaluation

The repository validates plugin structure, generated artifacts, and documentation. Contributors can run these checks from the repository root:

make generate-all
make validate STRICT=1
make garden

plugin-eval (opens in a new tab) also provides static skill checks without model calls. For example:

uv run --project plugins/plugin-eval plugin-eval score plugins/python-development/skills/python-testing-patterns --depth quick

The LLM judge and Monte Carlo layers are experimental and are not validated against human labels. Their scores describe model assessments and repeated simulations and do not prove that a task succeeds in a project. They require the optional llm dependencies. Model calls use ANTHROPIC_API_KEY when it is set and bill that API account. Otherwise they use your Claude Code login. See the evaluation guide (opens in a new tab) for setup and limits.

Quick start

Choose a plugin for the work you do:

See the plugin catalog (opens in a new tab) for all 94 plugins and the usage guide (opens in a new tab) for commands and examples.

Claude Code

Run these commands inside Claude Code to register the marketplace and install the Python development plugin:

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

Then ask Claude to use the Python agents for your task, or run the plugin’s scaffolding command:

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

Codex CLI

Run these commands in a terminal with a current Codex CLI:

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

The native Codex manifests expose source skills. The generated TOML agents and command-derived skills use a separate adapter path. Native installation also reads the original skill bodies, while generated copies split oversized bodies into reference files for the 8 KB budget. See the Codex setup details (opens in a new tab) before choosing an installation route.

Cursor

Cursor uses the committed plugin marketplace registry (opens in a new tab). Follow the Cursor setup instructions (opens in a new tab) to add the marketplace and install a plugin.

Skills only

Install an individual skill with GitHub CLI’s gh skill command or the npx skills installer. Select your coding tool when prompted. Both installers read skills from GitHub without cloning the repository and install them without the plugin’s agents, commands, or hooks.

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

gh skill requires GitHub CLI 2.90 or later. The npx command requires Node.js and npm. See skills-only installation (opens in a new tab) for other agents, install scopes, and version pinning.

Install from a clone

For OpenCode, Antigravity CLI, GitHub Copilot, and Pi, clone the repository and use the installer for your harness. Install uv (opens in a new tab), Python 3.12 or later, and the coding tool you want to use. The following clone command uses GitHub CLI:

gh repo clone wshobson/agents ~/agents
cd ~/agents

Run the command for your tool. Each target generates the artifacts and links all local plugins into that tool’s user configuration:

make install-opencode
make install-antigravity
make install-copilot
make install-pi

Pi’s agents require its reference subagent extension or a compatible extension. See the global install guide (opens in a new tab) for configuration paths and uninstall commands.

What’s inside

The agent, skill, and command counts cover local source files. External plugins provide their own components and installation requirements.

Documentation map

See CONTRIBUTING.md (opens in a new tab) to propose a change and authoring conventions (opens in a new tab) to write plugin content.

External memory integration

Pensyve (opens in a new tab) remains an optional external Claude Code marketplace entry. Major7 Apps maintains Pensyve, and this repository’s maintainer founded Major7 Apps. Its open-source runtime can run locally or on your own server. Pensyve Cloud closed on October 1, 2026 (opens in a new tab), and its dashboard, API, and hosted MCP endpoints are unavailable. Use the self-hosting guide (opens in a new tab) when configuring the runtime.

License

The repository is MIT licensed (opens in a new tab). External plugins have their own licenses.

Star history

(opens in a new tab)