Skip to content
csl
All skills

wshobson/agents

Meta

Plugin marketplace: 84 plugins, 192 subagents, 156 skills, 102 commands across architecture, languages, infra, security, ML.

by Seth Hobson (wshobson)wshobson/agents
Stars41.5k
Forks4.4k
Watchers319
Open issues6
LicenseMIT
Last push2 days ago

Install

install

$ /plugin marketplace add wshobson/agents

Update

update

$ /plugin marketplace update agents

Then run /reload-plugins (or restart Claude Code). Marketplaces re-sync at startup.

When to use it

  • You want a broad, curated library of production-ready subagents (language pros, security, infra, ML, SEO) instead of writing them yourself.
  • You need composable per-domain plugins where installing one loads only its agents/skills/commands into context.
  • You want multi-agent orchestrators for full-stack, security review, ML, or incident-response workflows.
  • You run the same agent definitions across Claude Code plus Codex, Cursor, OpenCode, Gemini, or Copilot from one source.

Commands & flags

/plugin marketplace add wshobson/agents
Register the marketplace in Claude Code.
/plugin install python-development
Install one of 84 single-purpose plugins (here: 3 Python agents + 16 skills).
/plugin install plugin-eval
Add the three-layer (static / LLM-judge / Monte Carlo) plugin quality evaluator.
uv run plugin-eval score path/to/skill --depth quick
Score a skill's structural and semantic quality with plugin-eval.
npx codex-marketplace add wshobson/agents
Install natively into Codex CLI (also Cursor/OpenCode/Gemini/Copilot supported).

How to use it

Add the marketplace, then install only the plugins you need (each is isolated and composable):

/plugin marketplace add wshobson/agents
/plugin install python-development      # one of 84 plugins

Browse the catalog with docs/plugins.md, docs/agents.md, and docs/agent-skills.md. Agents use a tiered model strategy (Opus for architecture/security, Sonnet for docs/testing, Haiku for fast ops). Evaluate plugin/skill quality with the bundled plugin-eval:

uv run plugin-eval certify path/to/skill

Non-Claude harnesses install via npx codex-marketplace add wshobson/agents (Codex/Cursor) or clone + make generate HARNESS=<name> (Gemini/OpenCode).

Why it earns its place

One of the most-starred Claude Code agent collections, by Seth Hobson (wshobson) — 192 domain-expert subagents, 156 skills, and 102 commands packaged as 84 granular, single-purpose plugins so you load only what you need. It ships a real tiered model strategy (Opus/Sonnet/Haiku per task), 16 multi-agent orchestrators, and a plugin-eval quality framework. The same Markdown source emits harness-native artifacts for Codex, Cursor, OpenCode, Gemini, and Copilot, so the library is portable beyond Claude Code.

Try it

claude

> Install the python-development and security plugins, then use the orchestrator to review this FastAPI service for vulnerabilities and missing tests.