Multi-Harness Agentic Plugin Marketplace — 84 Plugins, 192 Agents, 6 Harnesses From One Source
TL;DR
The wshobson/agents repository (36.7K stars, 4K forks) is a multi-harness agentic plugin marketplace — 84 plugins, 192 agents, 156 skills, and 102 commands built for Claude Code as the source of truth, then generated natively for OpenAI Codex CLI, Cursor, OpenCode, Gemini CLI, and GitHub Copilot. One plugins/ directory, six harness adapters, zero lowest-common-denominator translations.
The Cross-Harness Problem
By mid-2026, the AI coding agent landscape had fragmented into at least six serious harnesses. Claude Code, Codex CLI, Cursor, GitHub Copilot, Gemini CLI, and OpenCode each have their own plugin formats, skill manifests, agent definitions, and command registries. Teams that want portability face a stark choice: maintain six separate plugin repos, or accept lock-in to a single harness.
The wshobson/agents project solves this with a generative adapter pattern — one Markdown source, six native outputs [1].
What’s Inside
| Component | Count | Purpose |
|---|---|---|
| Plugins | 84 | Granular, single-purpose installable units (82 local + 2 external via git-subdir) |
| Agents | 192 | Domain experts (architecture, languages, infra, security, data, ML, docs, business, SEO) |
| Skills | 156 | Modular knowledge packages with progressive disclosure |
| Commands | 102 | Slash commands: scaffolding, security scans, test gen, infrastructure setup |
| Orchestrators | 16 | Multi-agent coordination workflows (full-stack, security, ML, incident response) |
Each plugin is isolated and composable. Installing a plugin loads only its components into context — not the entire marketplace. A python-development plugin might contain 3 agents, 1 command, and 16 skills, scoped to Python workflows only [1].
Architecture: Single Source, Native Output
The core innovation is the adapter generator pattern. The source of truth lives in plugins/ as Claude Code-native marketplace.json + plugin definitions. A make generate-all command produces harness-specific artifacts:
| Harness | Generated Output | Notes |
|---|---|---|
| Claude Code | marketplace.json + plugins/ (source of truth) |
Native format |
| Codex CLI | .agents/plugins/marketplace.json + plugins/*/.codex-plugin/plugin.json (committed); .codex/skills/, .codex/agents/ (gitignored) |
8 KB skill cap respected; commands mapped to skills |
| Cursor | .cursor-plugin/, .cursor/rules/ |
Thin marketplace + curated rules; reuses .claude/ |
| OpenCode | .opencode/agents/, .opencode/commands/, .opencode/skills/ |
permission: block from tools: allowlist |
| Gemini CLI | skills/, agents/, commands/ (TOML) |
Native skills + subagents per April 2026 spec |
| GitHub Copilot | .copilot/agents/, .copilot/skills/, .copilot/commands/ |
Markdown agent profiles; model maps to native Claude |
This is not a translation layer — each generator produces idiomatic artifacts for its target harness. The Codex CLI output respects the 8 KB skill file cap, OpenCode gets permission: blocks from tools: allowlists, and Gemini CLI gets TOML-format subagent definitions [1].
Model Tiers
The marketplace ships with a tiered model strategy that routes tasks to the right model by complexity:
| Tier | Model | Use Case |
|---|---|---|
| 0 | Fable 5 | Longest-horizon autonomous work — large migrations, multi-hour runs (opt-in, premium cost) |
| 1 | Opus | Architecture, security, code review, production-critical |
| 2 | inherit | User-chosen — backend, frontend, AI/ML, specialized |
| 3 | Sonnet | Docs, testing, debugging, API references |
| 4 | Haiku | Fast operational tasks, SEO, deployment, content |
Model aliases are refreshed June 2026 (Claude Fable 5 support added; Codex, Copilot, and OpenCode maps updated) [1].
Quick Start Setup
Claude Code (source of truth)
/plugin marketplace add wshobson/agents
/plugin install python-developmentAny of the 84 plugins can be installed the same way. The marketplace is auto-discovered from the committed marketplace.json.
Codex CLI and Cursor
npx codex-marketplace add wshobson/agents
# Cursor: add the marketplace, then /plugin install <name>Both read from committed registries in the repo.
Gemini CLI and OpenCode
gh repo clone wshobson/agents ~/agents && cd ~/agents
make generate HARNESS=gemini && gemini extensions install .
make install-opencodeThe clone + generate pattern produces harness-native artifacts without manual translation [1].
Quality Framework
The project includes plugin-eval, a three-layer evaluation system:
- Static analysis — deterministic structural checks (<2s, free)
- LLM Judge — semantic evaluation across 4 dimensions (~30s, using Haiku + Sonnet)
- Monte Carlo — statistical reliability via 50-100 simulated runs (~2-5 min)
uv run plugin-eval score path/to/skill --depth quick
uv run plugin-eval certify path/to/skillThe build pipeline also includes make validate for structural integrity and make garden for drift and dead-link detection [1].
Why This Matters for the Agent Ecosystem
Cross-harness portability is the next frontier for AI coding agents. The 2025 era of “pick one harness and build everything in it” is giving way to a multi-harness workflow where developers use Claude Code for architecture, Codex CLI for rapid iteration, and Cursor for in-editor polish — all pulling from the same skills and agent definitions [2][3].
The Firecrawl 2026 coding agent comparison notes that the frontier models have converged, so “the agent wrapper now decides your experience” [2]. When the models are largely interchangeable, the value shifts to the ecosystem of plugins, skills, and agents that surround them. A project like wshobson/agents makes that ecosystem portable.
The Agensi 2026 comparison confirms that all major agents now support SKILL.md, making skill portability a reality [3]. The wshobson/agents project extends this principle from individual skills to entire plugin marketplaces.
Related Reads
- What VS Code’s Coding Harness Teaches About Agent Evaluation — CodeIntel
- Claude Code vs OpenAI Codex CLI (2026): The Definitive Comparison — ToolBrain
- Claude Code MCP Integration Guide: Connecting Your Agent to Any Tool or API — ToolBrain
- Cursor vs Claude Code vs GitHub Copilot — Which AI Coding Tool Should You Use in 2026? — ToolBrain
References
[1] wshobson/agents — Multi-harness agentic plugin marketplace. GitHub. https://github.com/wshobson/agents
[2] Hiba Fathima, “Best AI Coding Agents in 2026: Harness, Cost, and Accuracy Compared.” Firecrawl Blog, June 2026. https://www.firecrawl.dev/blog/best-ai-coding-agents
[3] “Claude Code vs Cursor vs Codex CLI Compared (2026).” Agensi.io. https://www.agensi.io/learn/ai-coding-tools-comparison-2026
← Back to all posts
