Why it matters
This project simplifies AI agent development by enabling builders to create and deploy plugins once, with native support across multiple popular AI coding environments. Developers can leverage a vast library of pre-built agents and skills, reducing redundant work and accelerating the creation of complex agentic workflows.

What changed

The wshobson/agents repository introduces a multi-harness agentic plugin marketplace designed to serve as a single source of truth for AI agent components. This marketplace aggregates 94 granular plugins, 202 distinct agents categorized by domain expertise (e.g., architecture, security, ML), 183 modular skills with progressive disclosure, 105 commands for scaffolding and utility tasks, and 16 orchestrators for multi-agent coordination. The core innovation lies in its ability to generate idiomatic, harness-native artifacts for seven different agentic environments from a single Markdown source. These environments include Claude Code, Codex CLI, Cursor, OpenCode, Antigravity CLI, GitHub Copilot, and Pi. Each plugin is designed for isolation and composability, meaning only the necessary components are loaded into context upon installation. For instance, a python-development plugin might contain three Python agents, one scaffolding command, and 16 specialized skills, all auto-discovered from its directory structure. The project also outlines a tiered model strategy, mapping different AI models (Fable 5, Opus, Sonnet, Haiku) to specific use cases ranging from long-horizon autonomous work to fast operational tasks.

Why it matters for builders

For AI builders, this marketplace significantly streamlines the development and deployment of agentic workflows. Instead of adapting plugins for each individual AI coding environment, developers can now author their components once and have them natively integrated across multiple platforms. This reduces the overhead of maintaining separate codebases or translation layers. The extensive catalog of pre-built agents, skills, and commands allows developers to quickly assemble complex functionalities, accelerating prototyping and production cycles. Furthermore, the clear documentation on authoring and harness-specific capabilities empowers builders to contribute and integrate their own tools effectively.

Practical impact

Developers can begin using the wshobson/agents marketplace immediately. For Claude Code users, installation is as simple as /plugin marketplace add wshobson/agents followed by /plugin install [plugin-name]. Codex CLI and Cursor users can install the marketplace natively via npx codex-marketplace add wshobson/agents or by adding the marketplace within Cursor and then installing plugins. For environments like Antigravity CLI, OpenCode, and Pi, a clone-and-generate process is required using make commands. Individual skills can be installed directly into any agent using gh skill install wshobson/agents or npx skills add wshobson/agents, bypassing the need to clone the entire repository. Builders are encouraged to explore the documentation, particularly docs/harnesses.md for a detailed capability matrix and per-harness specifics, and docs/plugins.md and docs/agents.md for the full catalog.

Caveats and source limits

The source provides detailed information on the structure, content, and installation methods for the wshobson/agents marketplace across various harnesses. However, specific performance benchmarks or independent validation of the LLM judge and Monte Carlo quality evaluation methods are not detailed. The source also mentions premium costs for the Fable 5 model but does not specify pricing for other models or services. While the repository is marked as production-ready, the experimental nature of some quality evaluation tools should be noted by users. The integration details for external services like Pensyve and HOL Guard are provided, but their current operational status and any associated costs are not elaborated upon.

Sources

Written with AI assistance from the linked sources; every claim below was checked against them automatically. How we produce articles.

Claim check: 6/6 supported claims - 6 evidence links - 100% avg confidence
  • The wshobson/agents repository provides a multi-harness agentic plugin marketplace.supported - github.com
  • The marketplace supports seven agentic harnesses: Claude Code, Codex CLI, Cursor, OpenCode, Antigravity CLI, GitHub Copilot, and Pi.supported - github.com
  • The marketplace contains 94 plugins, 202 agents, 183 skills, and 105 commands.supported - github.com
  • Components are generated from a single Markdown source for native consumption by each harness.supported - github.com
  • The project utilizes a tiered model strategy mapping AI models like Fable 5, Opus, Sonnet, and Haiku to specific use cases.supported - github.com
  • Individual skills can be installed directly into any agent using `gh skill install` or `npx skills add`.supported - github.com

Caveats

  • Single-source caution: verify critical details at the linked source.
Radar score 78/100 - how it was calculated
Reliability82
Freshness8
Novelty74
Technical89
Developer96
Ecosystem66
Confidence100
  • Reliability 82: GitHub metadata supports source trust
  • Freshness 8: Fresh GitHub activity
  • Novelty 74: Novelty blends source metadata and enrichment
  • Technical 89: Repository technical metadata
  • Developer 96: Developer tooling signals
  • Ecosystem 66: Developer-oriented GitHub signal
  • Confidence 100: Claims have reliable evidence
Share
XLinkedInHacker News

Discussion

Loading comments...

Related articles

AI Coding - Oct 11, 2026Pojia-Next v8.6: Unified AI Client Prompt ManagementPojia-Next has released version v8.6, offering a zero-dependency Python script to manage prompts and personas across six AI clients: DSH, WorkBuddy, ZCode, Codex, Cursor, and Claude. The tool provides one-click operation, automatic backups, and a standalone executable.AI Coding - Oct 8, 2026PiG (Pi in Go) v0.4.1 ReleasedPiG (Pi in Go) has released version v0.4.1, offering a Go implementation of the Pi coding agent. This project aims for parity with the original TypeScript codebase, focusing on faithful translation rather than a rewrite.AI Coding - Oct 12, 2026Vetto v0.6.3: Daemon-less Sandbox for AI Coding AgentsVetto, a daemon-less OS security boundary and sandbox for AI coding agents, has released version v0.6.3. Written in Rust, it aims to provide a secure environment for tools like Codex, Claude Code, Cursor, and Aider.AI Coding - Oct 10, 2026AI DevKit: Control Plane for AI Coding AgentsAI DevKit is a TypeScript project serving as a control plane for AI coding agents. It recently released version cli@0.69.0, indicating active development and a focus on developer tools.AI Coding - Oct 11, 2026Claudexor v3.22.0: Multi-Model AI Control PlaneClaudexor has released version 3.22.0, a multi-harness control plane for AI coding assistants like Claude Code, Codex, Cursor, and OpenCode. It offers quota-aware subscription rotation, shared thread context, and cross-model review capabilities.AI Coding - Oct 11, 2026Gograph v1.7.11: Local Go Static Analysis for AI AgentsGograph, a local-only Go static analysis engine, has released version v1.7.11. It provides AI coding agents with deterministic structural awareness, including call graphs and impact analysis, via a built-in MCP server.