agentskills.io

Overview

概述

agentskills.io is the leading open standard for agent skill interoperability, with 20+ platforms, 100K+ installs, and 500+ published skills. It provides the specification, SDK, and public registry for cross-platform skill distribution.

agentskills.io 是 Agent Skill 互操作性的领先开源标准，拥有 20+ 平台支持、100K+ 安装量以及 500+ 已发布的 Skill。它为跨平台 Skill 分发提供了规范、SDK 和公共注册中心。

Details

详情

Mission: “Write once, run everywhere” for agent skills

**使命**：Agent 技能的“一次编写，到处运行”

Key Components

• Specification: JSON schema for skill manifests, execution protocol, registry API
• skills-ref SDK: Reference implementations in Python, TypeScript, Go
• Public registry: Centralized catalog of open-source skills
• Community governance: Technical steering committee, open contribution model

**核心组件** - **Specification**：技能清单的 JSON schema、执行协议、注册表 API - **skills-ref SDK**：Python、TypeScript 和 Go 语言的参考实现 - **Public registry**：开源技能的集中式目录 - **Community governance**：技术指导委员会、开放贡献模式

Adoption Metrics (Q1 2026)

• 20+ platforms supporting the standard
• 100K+ total skill installs across platforms
• 500+ published skills in public registry
• 15+ enterprise private registries deployed

**采用率指标（2026 年第一季度）** * 20+ 个平台支持该标准 * 跨平台累计 Skill 安装量 100K+ * 公共 Registry 中发布 500+ 个 Skill * 15+ 个企业私有 Registry 已部署

Supported Platforms

• Claude Code
• OpenClaw
• Windsurf
• Cursor
• Cline
• Continue
• Aider
• 13+ others

**支持的平台** - Claude Code - OpenClaw - Windsurf - Continue - 及其他 13+ 个平台

Skill Manifest Format

**Skill Manifest 格式**

```json { "name": "github-pr-review", "version": "2.1.0", "description": "Review GitHub pull requests with AI analysis", "permissions": ["github:read", "github:write"], "dependencies": ["github-api@^3.0.0"], "platforms": ["claude-code", "openclaw"], "author": "community", "license": "MIT" } ```

Registry API

• GET /api/skills - List all skills
• GET /api/skills/{name} - Get skill details
• GET /api/skills/{name}/versions - List versions
• POST /api/skills - Publish new skill (requires auth)
• DELETE /api/skills/{name} - Unpublish skill

**Registry API** - `GET /api/skills` - 列出所有 skills - `GET /api/skills/{name}` - 获取 skill 详情 - `GET /api/skills/{name}/versions` - 列出版本 - `POST /api/skills` - 发布新 skill (需要认证) - `DELETE /api/skills/{name}` - 取消发布 skill

Security Model

• Explicit permission declarations
• User approval flows for sensitive permissions
• Sandboxed execution
• Audit hooks for compliance

**安全模型** - 显式权限声明 - 敏感权限的用户审批流程 - 沙箱化执行 - 用于合规的审计钩子

China Adoption

• Used by domestic platforms (Coze, DingTalk AI)
• Private registries hosted on Alibaba Cloud, Tencent Cloud
• Compliance with MLPS 2.0 through audit logging
• Localized documentation and examples

**中国落地应用** - 被国内平台（Coze、钉钉 AI）采用 - 私有 Registry 托管于阿里云、腾讯云 - 通过审计日志满足 MLPS 2.0 合规要求 - 本地化的文档与示例

Competitive Landscape

• MCP (Model Context Protocol): Anthropic’s tool protocol, higher token overhead
• LangChain Tools: Python-specific, not cross-platform
• OpenAI Function Calling: OpenAI-specific, not portable

**竞争格局** - **MCP (Model Context Protocol)**：Anthropic 的工具协议，Token 开销较高 - **LangChain Tools**：仅支持 Python，不具备跨平台能力 - **OpenAI Function Calling**：仅限 OpenAI 使用，不可移植

Advantages

• Platform-agnostic: Skills work across 20+ platforms
• Open governance: Community-driven, not vendor-controlled
• Low overhead: Minimal token usage compared to MCP
• Proven scale: 100K+ installs demonstrate viability

**优势** - 平台无关：Skills 可在 20 多个平台上通用 - 开放治理：社区驱动，不受供应商控制 - 低开销：与 MCP 相比，Token 占用极少 - 经过验证的规模：10 万+ 次安装证明了其可行性

Connections

连接

• Related to: Skill Registry, Skill Factory Layers
• Mentioned in: agentskills.io Analysis, Skill Factory Framework

- 相关内容：[[enterprise-agent-china/concepts/skill-registry|Skill Registry]], [[enterprise-agent-china/concepts/skill-factory-layers|Skill Factory Layers]] - 提及于：[[enterprise-agent-china/sources/agentskills-io-analysis|agentskills.io Analysis]], [[enterprise-agent-china/sources/skill-factory-framework|Skill Factory Framework]]