wshobson/agents

wshobson/agents – Plug‑in Marketplace for AI‑powered Coding Assistants

What it is – A curated collection of plugins, agents, skills and commands that can be loaded into a variety of AI‑coding environments (Claude Code, OpenAI Codex CLI, Cursor, OpenCode, Antigravity CLI, GitHub Copilot, Pi). Each plugin groups Markdown‑defined instructions that tell an LLM how to perform a concrete development task – e.g. scaffold a FastAPI service, run a security scan, or apply a testing pattern.

Key pieces

Component	How it’s used
Plugins (94)	Top‑level bundles (e.g. python-development, security‑scanning). Installing a plugin makes all its agents, skills and commands available to the chosen harness.
Agents (202)	Small “sub‑agents” with a model alias (opus, sonnet, haiku, etc.) that carry out delegated work such as code review or dependency analysis.
Skills (184)	Re‑usable knowledge snippets or prompts that an agent can load when the situation matches (e.g. python‑testing‑patterns).
Commands (105)	Named CLI‑style actions you invoke (python-development:python-scaffold …) that may translate into a skill or a prompt template.

Supported AI coding tools (harnesses) – The repo ships adapters that translate the same Markdown source into the format each tool expects. Installation is either through the tool’s marketplace (Claude Code, Cursor) or via generated artifacts (make install‑codex, make install‑copilot, etc.).

Typical workflow

1. Add the marketplace – e.g. codex plugin marketplace add wshobson/agents.
2. Install a plugin – e.g. codex plugin add python-development@claude-code-workflows.
3. Run a command – e.g. python-development:python-scaffold Create a FastAPI service with tests. The harness loads the appropriate agent, injects the relevant skill, and the LLM (Claude, OpenAI, etc.) generates the code.

Why it matters – It lets teams standardise the “prompt engineering” that powers AI‑assisted development. Instead of each developer writing their own Claude or Codex prompts, they can share vetted, version‑controlled agents/skills that are automatically adapted to the tool they use.

Quality checks – make generate-all, make validate STRICT=1, and the plugin‑eval utility provide static validation and an experimental LLM‑based scoring of skills. Scores are produced by an LLM judge and Monte‑Carlo simulations; they are not human‑verified guarantees.

Getting started – Clone the repo, install uv and Python 3.12+, then run the appropriate make install‑<harness> target. For a single skill you can also use gh skill install … or npx skills add … without cloning.

License – MIT (plugins may carry their own licenses).

---

All details are taken directly from the repository’s README; no additional features are inferred.