AgentCommunity★ 38k

wshobson Agents

Updated Jul 5, 2026

An agentic plugin marketplace with 90 plugins, 199 agents, 161 skills, 106 commands, and 16 multi-agent orchestrators. Single Markdown source builds to five harnesses: Claude Code, Codex CLI, Cursor, OpenCode, and Gemini CLI. Includes a tiered model strategy (Fable/Opus/Sonnet/Haiku) and a three-layer plugin-eval quality framework.

View on GitHub

What it does

/plugin marketplace add wshobson/agents
Register the wshobson plugin marketplace in Claude Code.
/plugin install python-development
Install the python-development plugin (3 agents: python-pro, django-pro, fastapi-pro; 16 skills; 1 command).
make generate-all
Build all five harness-native artifact sets (Claude Code, Codex, Cursor, OpenCode, Gemini) from the single Markdown source.
uv run plugin-eval certify path/to/skill
Run the three-layer evaluation (static + LLM judge + Monte Carlo) to certify a plugin for quality.
orchestrator plugins (16)
Multi-agent coordination workflows for full-stack, security, ML, and incident response scenarios.

Files (1)

OVERVIEW.mdprimary · markdown · 2.0 KB

## What it does

An agentic plugin marketplace with 90 plugins, 199 agents, 161 skills, 106 commands, and 16 orchestrators — all from a single Markdown source that builds to five harnesses natively. Installing a plugin loads only its components into context, not the entire marketplace.

**What each plugin contains:**

Each plugin is a directory with a plugin.json manifest, agents/, commands/, and skills/ subdirectories. Example: the python-development plugin ships 3 Python agents (python-pro, django-pro, fastapi-pro), 1 scaffolding command, and 16 specialized skills covering async, testing, packaging, and more.

**Tiered model strategy:** Each agent specifies a model in frontmatter. Fable 5 for longest-horizon autonomous work, Opus for architecture/security/code review, Sonnet for everyday coding, Haiku for fast operational tasks like SEO and deployment.

**Multi-harness support:** Claude Code is the source of truth. Codex and Cursor install from committed registries. Gemini and OpenCode install via clone + make generate. A `make generate-all` command builds all five harnesses from the single Markdown source.

**Plugin eval framework:** Three-layer evaluation — static structural analysis (<2s), LLM Judge across 4 dimensions (~30s), and Monte Carlo reliability via 50–100 simulated runs (~2–5 min). Run `uv run plugin-eval certify path/to/skill` to certify a plugin.

## Install

**Claude Code:**

```
/plugin marketplace add wshobson/agents
/plugin install python-development
/plugin install security-scanning
/plugin install ml-engineering
```

**Codex CLI:**

```bash
npx codex-marketplace add wshobson/agents
```

**Gemini / OpenCode (via clone):**

```bash
gh repo clone wshobson/agents ~/agents && cd ~/agents
make generate HARNESS=gemini && gemini extensions install .
```

## When to use

When you want composable, independently installable agents with a proven quality-eval pipeline and a model routing strategy already baked in. Install only the domain plugins you need.