什么是 能力感知系统？

能力感知系统解决了智能体对自定义安装缺乏了解的问题。通过实施技能优先的方法，它为智能体提供了一种结构化的方式来查看可用能力的更高层级描述，并且仅在特定任务需要时才加载完整的文档。这种架构确保了 Openclaw 技能始终是可发现的，而不会在不使用时让不必要的文档细节堆满智能体提示词。

该系统是构建强大本地生态系统的基础层，智能体可以通过在正确的时间引入正确的工具来处理复杂的工作流。它弥补了“安装了工具”与“智能体真正知道如何以及何时高效使用工具”之间的差距。

下载入口:https://github.com/openclaw/skills/tree/main/skills/pfaria32/capability-awareness

安装与下载

1. ClawHub CLI

从源直接安装技能的最快方式。

2. 手动安装

将技能文件夹复制到以下位置之一

优先级：工作区 > 本地 > 内置

3. 提示词安装

将此提示词复制到 OpenClaw 即可自动安装。

能力感知系统 应用场景

• 使智能体能够自我发现添加到工作区的各种新能力。
• 通过仅加载相关 Openclaw 技能的文档来优化 Token 使用。
• 在不同的开发团队之间标准化技能文档格式。
• 为智能体工具集和能力提供一个中心化的注册表。

1. 为智能体提供一份包含可用技能简短摘要的能力卡列表。
2. 当收到用户请求时，智能体扫描这些描述，以查看是否有任何 Openclaw 技能适用于当前上下文。
3. 如果识别出匹配项，智能体执行读取命令以访问位于技能目录中的完整 SKILL.md 文件。
4. 智能体将文档中的特定指令集成到其当前的推理链中。
5. 智能体根据新获得的技能知识执行所需的操作。

能力感知系统 配置指南

要安装该系统，请导航到您的工作区并克隆仓库：

cd /home/node/.openclaw/workspace
git clone https://github.com/pfaria32/openclaw-capability-awareness.git projects/capability-awareness-system

确保您开发的任何新 Openclaw 技能都放置在 workspace/skills/ 目录中，并配有格式正确的 SKILL.md 文件供智能体读取。

能力感知系统 数据架构与分类体系

Capability Awareness System

Makes OpenClaw agents aware of custom skills and capabilities. Skills-first approach with on-demand documentation loading. Zero overhead when not in use. Foundation for skill marketplace discovery.

When to use: Skill discovery, capability documentation, agent self-awareness of available tools

What to know:

Repository: https://github.com/pfaria32/openclaw-capability-awareness

Problem

Default OpenClaw agents don't know about custom skills you've installed. They need to:

1. Discover what capabilities exist
2. Know when to use each skill
3. Access skill documentation on demand

Solution

Skills-First Approach:

• Agent sees skill descriptions in prompt
• Reads full SKILL.md when topic is relevant
• Zero overhead when skills not in use
• Simple, proven, low-risk

Implementation Options

Option 1: Skills-First (Recommended for v1)

Add capability cards to agent prompt:

Available Skills:
- token-economy: Model routing and cost optimization
- health-tracking: Apple Health and Strava integration
- memory-system: RAG-based semantic search

Agent reads full SKILL.md when needed.

Option 2: Full Injection (Advanced)

• Router-gated skill loading
• Dynamic prompt injection
• Context-aware capability exposure
• Zero baseline cost (only loads when relevant)

Installation

cd /home/node/.openclaw/workspace
git clone https://github.com/pfaria32/openclaw-capability-awareness.git projects/capability-awareness-system

Usage

Current Implementation (Skills-First)

Skills are documented in workspace/skills/*/SKILL.md. The agent loads these automatically through the AGENTS.md workflow:

## Skills (mandatory)
Before replying: scan   entries.
- If exactly one skill clearly applies: read its SKILL.md at  with `read`, then follow it.

This is already working! Just add new skills to the workspace/skills directory.

Future Implementation (Full Injection)

See repository for:

• Router design and schema
• Embedding-based skill matching
• Dynamic prompt injection strategy
• Cost/token analysis

Status

? Skills-First approach — Deployed and working ?? Full Injection design — Documented, not yet implemented

Attribution

Built to support the emerging OpenClaw skill ecosystem. Simple beats clever.

Documentation

Implementation options, design decisions, and upgrade path documented in repository.