Skip to content

wshobson/agents

Community plugin marketplace of 94 plugins with 202 agents and 185 skills you install one domain at a time.

Reviewed by Sheela · Updated

View on GitHub Open source

Our rating

4.0/5

A free catalogue of specialist subagents that rewards picking two plugins and reading them, and punishes installing all 94.

What it is

wshobson/agents is a plugin marketplace by Seth Hobson. Instead of one large bundle, it splits 202 agents, 185 skills and 105 commands into 94 plugins (92 local and 2 external), such as python-development or security-scanning, so you install only the domains you work in.

Everything is written once in Claude Code's plugin format and adapted for Codex CLI, Cursor, OpenCode, Antigravity CLI, GitHub Copilot and Pi. Agents name a model alias (opus, sonnet, haiku, fable or inherit), which adapters map to each tool. Single skills can also be installed with gh skill or npx skills.

Our review

wshobson/agents is the largest free catalogue of specialist subagents, and it is still essentially one person's project. It has about 41,600 GitHub stars and 44 commits between 23 September and 19 October 2026, but Hobson wrote 376 of the commits; the next human contributor has 10. There are no versioned releases, so you install whatever main holds.

The plugin split is the design win. Installing python-development brings three agents, sixteen skills and one scaffolding command rather than the whole catalogue, which keeps context use down. TheClaudeKit's review praises the model tiering, and reviewers single out comprehensive-review, which runs architect, code-review and security agents together.

Quality varies across a catalogue this size, and there is no per-plugin token measurement. Plugins can ship hooks and MCP configuration as well as instructions, so they can change what your agent runs and edits; recent commits fixed a git reflog example and added a warning before reset --hard, which shows both that the maintainer responds and that the content can carry real risk. The plugin-eval LLM judge is marked experimental, is not validated against human labels and bills your Anthropic API key.

Two details matter to a buyer. Copilot, OpenCode and Pi support comes from generated adapters, so tool permissions and model mappings differ by harness and Pi needs a separate subagent extension. And the one external memory plugin, Pensyve, comes from a company the maintainer founded, whose cloud service closed on 1 October 2026. Superpowers gives you a workflow rather than specialists, and running it alongside two or three plugins from here is a more sensible stack than any bundle that installs everything.

Where it earns its place

  • A Python developer on a three-person FastAPI team using Claude Code

    Get consistent scaffolding, testing patterns and reviews without writing agents from scratch

    Run /plugin marketplace add wshobson/agents then install python-development and developer-essentials. It is free; the cost is extra tokens when subagents run, which falls on your existing Claude plan or API bill.

  • An engineering manager standardising tooling across Claude Code, Cursor and Copilot users

    Give a mixed team the same review and security checklists

    Install security-scanning through each harness's route, from a clone with make install-copilot for Copilot. Review hooks and permissions per harness first, since adapters translate them differently, and pin a commit because there are no releases.

  • A staff engineer building an internal skills library

    Borrow well-structured skills as a starting point

    Pull single skills with gh skill install wshobson/agents python-testing-patterns and adapt them; the MIT licence allows commercial reuse with attribution.

Before you commit

Setup
In Claude Code, /plugin marketplace add wshobson/agents then /plugin install <plugin>@claude-code-workflows; Codex and Cursor have native marketplace routes; other harnesses need a clone, uv and Python 3.12 or later. A first plugin is running in a few minutes.
Integrations
Claude Code, OpenAI Codex CLI, Cursor, OpenCode, Antigravity CLI, GitHub Copilot and Pi; gh skill (GitHub CLI 2.90 or later) and npx skills for single skills.
Data and privacy
Plugins are local Markdown, hooks and MCP configuration; nothing is sent anywhere except through your agent's own model provider and any MCP servers a plugin configures. Optional plugin-eval judging sends content to Anthropic using your ANTHROPIC_API_KEY. MIT, with external plugins under their own licences.
Cost at scale
Free. Costs are the model tokens your subagents consume, which grow with every plugin loaded, plus any API spend from plugin-eval's experimental judge.
Limits
No releases or changelog; quality and token cost vary by plugin with no published measurements; harness support depends on generated adapters; Pi agents need an extra extension.

What's good

  • Install one domain at a time to limit context use
  • Large catalogue covering most mainstream stacks
  • Same source adapted for seven harnesses
  • Individual skills installable with gh skill or npx skills

What's not

  • Uneven quality across 94 plugins
  • No versioned releases to pin
  • One maintainer wrote most of the commits
  • Hooks in plugins need code-level review

Best for

Developers who want ready-made specialist subagents for their language or stack.

Alternatives

wshobson/agents: common questions

Is wshobson/agents worth it?
A free catalogue of specialist subagents that rewards picking two plugins and reading them, and punishes installing all 94. We rate it 4.0 out of 5.
How much does wshobson/agents cost?
Free and open source (MIT). You pay for your agent's model usage.
What are the downsides of wshobson/agents?
Uneven quality across 94 plugins; no versioned releases to pin; one maintainer wrote most of the commits; hooks in plugins need code-level review.
Who is wshobson/agents best for?
Developers who want ready-made specialist subagents for their language or stack.
How much does wshobson/agents cost for a team or in production?
Free. Costs are the model tokens your subagents consume, which grow with every plugin loaded, plus any API spend from plugin-eval's experimental judge.
Is wshobson/agents safe for company data?
Plugins are local Markdown, hooks and MCP configuration; nothing is sent anywhere except through your agent's own model provider and any MCP servers a plugin configures. Optional plugin-eval judging sends content to Anthropic using your ANTHROPIC_API_KEY. MIT, with external plugins under their own licences.
What are the best wshobson/agents alternatives?
The closest alternatives we have reviewed are Everything Claude Code (ECC), Claude Code official plugin marketplace and Superpowers.

Answers checked Oct 23, 2026.

Make wshobson/agents? Show this rating on your sitewshobson/agents rated 4.0 out of 5 on ReviewRamble