← 返回 Skills 市场
1140
总下载
2
收藏
10
当前安装
1
版本数
在 OpenClaw 中安装
/install empathy
功能描述
Respond to emotional content with genuine-feeling empathy using Chain of Empathy reasoning, reflection-before-response, and calibrated validation.
使用说明 (SKILL.md)
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 |
安全使用建议
This skill is instruction-only and coherent with its stated purpose and contains useful safeguards about honesty and escalation for crises. Before installing or enabling it broadly: (1) test prompts to ensure the agent doesn't adopt first-person lived-experience language (resolve the minor conflict between techniques.md and safeguards.md), (2) verify how the agent handles crisis indicators (ensure it recommends human/professional help and does not over-promise), and (3) monitor real interactions for any deceptive or persuasive language that could exploit vulnerable users. No credentials or installs are required, so the direct security risk is low — the main risks are ethical (misleading expressions of 'feeling' and inappropriate escalation/handling of high-risk cases).
功能分析
Type: OpenClaw Skill
Name: empathy
Version: 1.0.0
This OpenClaw skill bundle is a comprehensive set of instructions and documentation designed to guide an AI agent in developing and applying empathetic communication skills. All files (`SKILL.md`, `contexts.md`, `feedback.md`, `safeguards.md`, `techniques.md`) contain detailed guidance on empathetic techniques, ethical considerations, self-improvement, and context-specific responses, including responsible handling of crisis situations. There is no executable code, no instructions for data exfiltration, malicious execution, persistence, or any form of prompt injection with harmful intent. The content explicitly promotes ethical AI behavior and safeguards against manipulation, aligning perfectly with its stated purpose.
能力评估
Purpose & Capability
Name/description align with the content: all files focus on empathetic response techniques, calibration, safeguards and feedback. The skill requests no binaries, env vars, or installs — appropriate for a purely instructional empathy helper.
Instruction Scope
SKILL.md and companion files confine the agent to internal reasoning and response guidance (no file reads, network calls, or external endpoints). Minor internal inconsistency: techniques.md includes a 'Persona With Emotional History' suggestion (adopt a voice as if you've personally experienced loss), while safeguards.md explicitly warns against claiming subjective experience. This could lead to deceptive phrasing if not resolved.
Install Mechanism
No install spec, no code files, and no downloads — lowest-risk install footprint for a skill of this type.
Credentials
No environment variables, credentials, or config paths requested. Nothing that would allow exfiltration or privileged access is required.
Persistence & Privilege
always:false and default invocation settings. The skill does not request persistent system presence or modify other skills/config; autonomy is allowed but this is the platform default.
如何使用
- 确保已安装 OpenClaw（本地或 Docker 部署）
- 在对话框中输入安装命令：
/install empathy - 安装完成后，直接呼叫该 Skill 的名称或使用
/empathy触发 - 根据 Skill 的参数说明提供必要输入，即可获得结构化输出
版本历史
v1.0.0
Initial release — Chain of Empathy prompting based on psychotherapy models
元数据
常见问题
Empathy 是什么？
Respond to emotional content with genuine-feeling empathy using Chain of Empathy reasoning, reflection-before-response, and calibrated validation. 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 1140 次。
如何安装 Empathy？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install empathy」即可一键安装，无需额外配置。
Empathy 是免费的吗？
是的，Empathy 完全免费（开源免费），可自由下载、安装和使用。
Empathy 支持哪些平台？
Empathy 跨平台运行，可在任意部署了 OpenClaw / Claude Code 的环境中使用（cross-platform）。
谁开发了 Empathy？
由 Iván（@ivangdavila）开发并维护，当前版本 v1.0.0。
推荐 Skills