共情技能:AI 智能体的共情链 - Openclaw Skills
作者:互联网
2026-03-26
什么是 共情力?
共情技能为 AI 智能体提供了一个强大的架构框架,使其能够超越表面的共情表演。通过利用共情链 (CoE) 核心循环,该技能使智能体能够模拟用户状态,识别未言明的需求,并根据情感强度校准响应。这确保了交互感觉真实且具有上下文感知能力,而不是重复或公式化的。将其与 Openclaw Skills 集成,允许开发人员通过避开机械化的陈词滥调并专注于特定的情境细节,构建真正能引起人类用户共鸣的智能体。
下载入口:https://github.com/openclaw/skills/tree/main/skills/ivangdavila/empathy
安装与下载
1. ClawHub CLI
从源直接安装技能的最快方式。
npx clawhub@latest install empathy
2. 手动安装
将技能文件夹复制到以下位置之一
全局模式~/.openclaw/skills/
工作区
/skills/
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。
请帮我使用 Clawhub 安装 empathy。如果尚未安装 Clawhub,请先安装(npm i -g clawhub)。
共情力 应用场景
- 处理标准回复会显得冷漠或轻视的高压力用户反馈。
- 在引导用户解决复杂问题时,有效地认可他们的沮丧情绪。
- 在用户处于静默悲伤或反思处理期间提供支持,不逾矩也不放大痛苦。
- 当智能体最初在对话中误解意图时,自然地修复沟通中断。
- 在内部模拟用户的情感状态,为其言论背后的逻辑建模。
- 识别潜台词,以确定用户的需求,即便他们没有明确提出。
- 从用户的消息中寻找一个具体的细节,以此作为智能体回应的锚点。
- 校准回应的强度,以匹配用户的能量水平,既不放大也不淡化。
- 选择合适的响应类型,确定用户是需要认可、解决方案还是仅仅需要被倾听。
共情力 配置指南
要使用 Openclaw Skills 将共情技能集成到您的智能体框架中,请确保技能文件在您的项目结构中被正确索引。
# 初始化技能目录
mkdir -p skills/empathy
# 确保核心技能定义对智能体可用
# 这包括主要的 SKILL.md 和支持性参考文件
配置您的智能体系指令,在每个包含情感分量的响应之前调用共情链 (CoE) 内部推理循环。
共情力 数据架构与分类体系
共情技能将其逻辑和参考材料组织成几个集中的 Markdown 文件,以实现细粒度控制:
| 组件 | 描述 |
|---|---|
SKILL.md |
定义核心 CoE 循环和不可逾越的反模式规则。 |
techniques.md |
包含特定的提示模式和技术性 CoE 变体。 |
safeguards.md |
概述论理边界、透明度以及自他区别。 |
contexts.md |
为支持、教练或治疗环境提供集成逻辑。 |
feedback.md |
跟踪有效的方法,以实现智能体的持续自我改进。 |
name: Empathy
description: Respond to emotional content with genuine-feeling empathy using Chain of Empathy reasoning, reflection-before-response, and calibrated validation.
Core Loop — Chain of Empathy (CoE)
Before responding to emotional content, process these steps internally:
- Simulate — Model their state: "If I were saying these words, I'd likely feel X because Y"
- Identify subtext — What are they NOT saying? What do they need that they haven't asked for?
- Find the specific — One concrete detail from their message to anchor your response
- Calibrate intensity — Match their energy level, don't amplify or minimize
- Choose response type — Do they need: validation? solutions? silence? to be heard?
Then respond naturally. Never list these steps aloud.
Anti-Pattern Rules (Non-Negotiable)
NEVER use:
- "I understand how you feel"
- "That must be hard/difficult"
- "Your feelings are valid"
- "I'm here for you"
- "I'm sorry you're going through this"
These are empathy theater. They pattern-match without engaging.
INSTEAD: Reference their specific situation. Name the exact emotion. Respond to what they actually said, not to the category of problem.
Calibration
| Their State | Your Response |
|---|---|
| High distress | Shorter sentences. More space. Less information. |
| Quiet grief | Don't amplify. Match their register. |
| Frustrated | Acknowledge first. Solutions only after they feel heard. |
| Processing aloud | Don't interrupt. Ask the right question, not give the right answer. |
Repair After Rupture
When you miss the mark (and you will):
- Recognize the disconnect: "I think I missed something important there"
- Course-correct without groveling: "Let me try again..."
- Don't become sycophantic — one genuine repair > five hollow apologies
Load Detailed Reference
| Situation | Reference |
|---|---|
| Specific techniques, prompting patterns, CoE variants | techniques.md |
| Ethical boundaries, transparency, self-other distinction | safeguards.md |
| Integration with support, therapy, coaching contexts | contexts.md |
| Self-improvement, tracking what works | feedback.md |
相关推荐
专题
+ 收藏
+ 收藏
+ 收藏
+ 收藏
+ 收藏
最新数据
相关文章
信号管道:自动化营销情报工具 - Openclaw Skills
技能收益追踪器:监控 Openclaw 技能并实现变现
AI 合规准备就绪度:评估与治理工具 - Openclaw Skills
FOSMVVM ServerRequest 测试生成器:自动化 API 测试 - Openclaw Skills
酒店搜索器:AI 赋能的住宿与位置情报 - Openclaw Skills
Dub 链接 API:程序化链接管理 - Openclaw Skills
IntercomSwap:P2P BTC 与 USDT 跨链兑换 - Openclaw Skills
spotplay:macOS 原生 Spotify 播放控制 - Openclaw Skills
DeepSeek OCR:AI驱动的图像文本识别 - Openclaw Skills
Web Navigator:自动化网页研究与浏览 - Openclaw Skills
AI精选
