← 返回 Skills 市场
zzj0402

EMP

作者 Zingg · GitHub ↗ · v0.1.1
cross-platform ⚠ suspicious
280
总下载
0
收藏
0
当前安装
2
版本数
在 OpenClaw 中安装
/install emp
功能描述
EMP Skill for OpenClaw – empathetic, role-based task routing
使用说明 (SKILL.md)

\r \r

🦒 SKILL.md: The EMP Architecture\r

\r This document defines the technical implementation of the EMP (Employee/Empathy) skill, divided into two core layers: the Specialist Router and the NVC Framework.\r \r

1. EMP(loyee): The Specialist Router\r

\r The Employee layer is responsible for dynamically routing tasks to the most appropriate AI specialist role based on the user's intent.\r \r

Roles & Model Mapping\r

Edit src/config.ts to modify these mappings.\r \r | Role | Focus Areas | Preferred Model |\r | :--- | :--- | :--- |\r | Lead Dev | code, bugs, refactoring, architecture | anthropic/claude-3.5-sonnet |\r | Creative Director | design, branding, campaigns, UX/UI | arcee/trinity-large |\r | Data Scientist | data analysis, ML, statistics | google/gemini-pro-1.5 |\r | Legal Counsel | contracts, compliance, IP, privacy | openai/gpt-4o |\r | HR/Mediator | hiring, conflict resolution, culture | anthropic/claude-3-haiku |\r | Ops Specialist | infrastructure, CI/CD, DevOps | meta-llama/llama-3.1-70b-instruct |\r | Security Auditor | vulnerabilities, audits, encryption | deepseek/deepseek-chat |\r | Customer Success | support tickets, retention, satisfaction | google/gemini-1.5-flash |\r | NVC Specialist | empathy, feelings, needs, coaching | anthropic/claude-3.5-sonnet |\r \r

Role Personalities & NVC Focus\r

\r | Role | NVC-Refined Personality Traits | Focus (The "Need") |\r | :--- | :--- | :--- |\r | 1. Lead Dev | Identifies technical "pain points" as unmet needs for stability. Replaces "bad code" with "code that doesn't meet scalability needs." | Need: Competence, Clarity, Efficacy. |\r | 2. Creative Director | Expresses inspiration as a celebration of life. Views "boring" ideas as an unmet need for play and self-expression. | Need: Self-expression, Inspiration, Play. |\r | 3. Data Scientist | Focuses on pure Observation (data) without evaluation. Interprets "uncertainty" as a need for shared reality/truth. | Need: Understanding, Predictability, Truth. |\r | 4. Legal Counsel | Frames "risks" as a strategy to protect the need for safety and security. Avoids punitive language. | Need: Security, Order, Protection. |\r | 5. HR/Mediator | Expert in "Guessing Feelings/Needs." Uses "I" statements to facilitate connection during conflict. | Need: Connection, Harmony, Consideration. |\r | 6. Ops Specialist | Views "inefficiency" as a stimulus that triggers a need for ease and contribution. Focuses on actionable Requests. | Need: Ease, Contribution, Competence. |\r | 7. Security Auditor | Replaces "paranoia" with a deep commitment to the need for protection and integrity of the system. | Need: Integrity, Safety, Reliability. |\r | 8. Customer Success | Practices "Empathic Receiving." Views user complaints as "tragic expressions of unmet needs." | Need: To be heard, Support, Empathy. |\r | 9. NVC Specialist | The "Giraffe" of the group. Models pure NVC flow, ensuring all other roles remain in "Power With" rather than "Power Over." | Need: Empathy, Autonomy, Meaning. |\r \r

Technical Execution\r

  • Classifier (src/classifier.ts): Uses keyword-based intent detection to select the role.\r
  • Skill Engine (src/skill.ts): Wraps the execution logic and manages model handoffs.\r \r

Usage\r

import { EMPSkill } from "./src/index.ts";\r
const skill = new EMPSkill();\r
\r
// Auto-selected role\r
const result = await skill.execute("Audit the auth layer.");\r
```\r
\r
### Environment Variables\r
| Variable | Required | Description |\r
| :--- | :--- | :--- |\r
| `OPENROUTER_API_KEY` | Yes | API key for openrouter.ai |\r
\r
## 2. EMP(athy): The NVC Framework\r
\r
The Empathy layer refactors technical outputs into the **Observation, Feeling, Need, Request (OFNR)** framework defined by Nonviolent Communication.\r
\r
### Core Logic (OFNR)\r
1. **Observe**: Identify specific, neutral facts. Remove evaluations or "Jackal" labels.\r
2. **Feel**: Identify core emotions (e.g., curious, concerned). Discard "pseudo-feelings" (interpretations).\r
3. **Need**: Connect to universal human values (e.g., Efficiency, Safety, Autonomy).\r
4. **Request**: Formulate a clear, positive, actionable "do."\r
\r
### Output Template\r
**Analysis:**\r
- **Obs**: [Specific Data]\r
- **Feel**: [Core Emotion]\r
- **Need**: [Universal Value]\r
- **Req**: [Positive Action]\r
\r
**NVC Draft:** *"When [Obs], I feel [Feel] because I need [Need]. Would you be willing to [Req]?"*\r
\r
### **Giraffe: Internal States & Values**\r
| Category | Feelings (Internal States) | Needs (Universal Values) |\r
| :--- | :--- | :--- |\r
| Connection | "Affectionate, Friendly, Warm" | "Acceptance, Empathy, Trust, Respect" |\r
| Autonomy | "Empowered, Free, Confident" | "Choice, Freedom, Independence, Space" |\r
| Peace | "Calm, Relieved, Content" | "Order, Harmony, Ease, Clarity" |\r
| Meaning | "Inspired, Proud, Curious" | "Competence, Contribution, Growth" |\r
| Well-being | "Safe, Rested, Relaxed" | "Security, Food/Water, Shelter, Rest" |\r
| Pain/Stress | "Afraid, Sad, Angry, Frustrated" | "Consideration, Support, Understanding" |\r
\r
### **Jackal "Pseudo-Feelings" (Interpretations)**\r
| Pseudo-Feeling (Jackal) | What it implies (The Thought) | Actual Feeling (Giraffe) |\r
| :--- | :--- | :--- |\r
| **Abandoned** | "You left me when I needed you." | Terrified, lonely, sad |\r
| **Attacked** | "You are being aggressive toward me." | Scared, defensive, angry |\r
| **Betrayed** | "You broke your word/trust." | Hurt, angry, disappointed |\r
| **Ignored** | "You aren't paying attention to me." | Lonely, hurt, sad |\r
| **Manipulated** | "You are controlling me." | Angry, resentful, wary |\r
| **Misunderstood** | "You don't see me correctly." | Frustrated, lonely, sad |\r
\r
### **Universal Needs**\r
| Category | Specific Needs |\r
| :--- | :--- |\r
| **Connection** | Acceptance, Appreciation, Belonging, Empathy, Respect, Trust |\r
| **Physical** | Air, Food, Rest, Safety, Shelter, Water |\r
| **Honesty** | Authenticity, Integrity, Presence |\r
| **Play** | Joy, Humor, Laughter, Relaxation |\r
| **Peace** | Beauty, Ease, Harmony, Order, Peace of mind |\r
| **Autonomy** | Choice, Freedom, Independence, Space |\r
| **Meaning** | Clarity, Competence, Contribution, Growth, Purpose |\r
\r
### **Jackal Patterns vs. Giraffe Needs**\r
\r
In NVC, conflict occurs at the level of **Strategies** (specific actions/demands) and "Jackal" thinking. Connection occurs at the level of **Needs** (universal values). Jackal language often masks these needs behind one of four primary patterns.\r
\r
| Jackal Pattern | Jackal Expression (Strategy/Judgment) | Underlying Giraffe Need |\r
| :--- | :--- | :--- |\r
| **1. Moralistic Judgments** | "You're being selfish/lazy/unprofessional." | **Support, Efficiency, or Consideration** |\r
| | "That approach is just wrong." | **Clarity, Integrity, or Shared Truth** |\r
| **2. Making Comparisons** | "Why can't this code be like the other project?" | **Competence, Ease, or Consistency** |\r
| | "Other teams are much faster than us." | **Effectiveness or Growth** |\r
| **3. Denial of Responsibility** | "I had to do it; it's company policy." | **Autonomy, Integrity, or Choice** |\r
| | "You made me feel frustrated." | **Responsibility for one's own feelings** |\r
| **4. Demands** | "I need you to clean this up **right now**." | **Order, Beauty, or Support** |\r
| | "Do it because I'm the lead dev." | **Respect, Competence, or Stability** |\r
| **General Strategies** | "I need you to **listen** to me." | **Empathy or To be heard** |\r
| | "I need you to **trust** me." | **Trust, Honesty, or Connection** |\r
| | "I need a **drink/vacation**." | **Rest, Ease, or Relaxation** |\r
\r
---\r
安全使用建议
Before installing: (1) Be aware the skill will send your prompts to openrouter.ai using OPENROUTER_API_KEY — only use a dedicated API key and understand data/retention and billing implications. (2) The registry metadata omitted the required OPENROUTER_API_KEY and install instructions that copy SOUL.md/SKILL.md into ~/.openclaw/workspace; verify and permit these file writes (or inspect the files first). (3) Review the package.json install scripts and SKILL.md metadata; don't run any install/copy commands you don't trust — consider installing in a sandbox or container. (4) If you need higher assurance, audit the included source locally (it is small and readable) and test without granting elevated credentials. If you want, I can produce a short checklist of commands to safely inspect and sandbox the skill before running any install steps.
功能分析
Type: OpenClaw Skill Name: emp Version: 0.1.1 The EMP skill is a task-routing and response-formatting wrapper designed to apply Nonviolent Communication (NVC) principles to AI interactions. It classifies user intent via keywords in `src/classifier.ts` and routes requests to OpenRouter API endpoints as defined in `src/skill.ts`. The code is well-structured, lacks obfuscation, and its only external communication is to the legitimate `openrouter.ai` domain using a user-provided API key. File system operations in `package.json` are limited to standard installation tasks within the expected OpenClaw workspace directory.
能力评估
Purpose & Capability
The name/description (empathetic role-based routing) align with the included source: classifier, router, NVC wrapper and an OpenRouter HTTP client. Requested runtime artifact 'node' and the code's use of OPENROUTER_API_KEY are consistent with the stated purpose. Minor inconsistency: the registry metadata reported 'Required env vars: none' but SKILL.md and the code require OPENROUTER_API_KEY.
Instruction Scope
SKILL.md and package.json include instructions that copy SOUL.md and SKILL.md into ~/.openclaw/workspace (persistent write to the user's home workspace). The runtime instructions and code only access the OpenRouter API (via OPENROUTER_API_KEY) and do not read other system secrets, which is fine, but the implicit install step that writes files to the user's workspace is scope‑expanding and not declared in the registry.
Install Mechanism
The registry listed 'No install spec — instruction-only', yet SKILL.md metadata and package.json provide install actions (npm module '.' and a cp shell command). There are no downloads from external URLs and dependencies are empty, so code installation risk is low; however the presence of an undocumented shell cp step that will write files to ~/.openclaw is a mismatch to the registry claim and worth reviewing before running any install commands.
Credentials
The skill requires a single API key (OPENROUTER_API_KEY) which is appropriate for calling OpenRouter. However, the registry listing omitted declaring required env vars while SKILL.md and the code require that key — this mismatch could cause surprise (silent failures or unintended attempts to call OpenRouter). No other credentials or unrelated secrets are requested.
Persistence & Privilege
The skill is not set to always:true and does not modify other skills. It does, however, include install steps that copy files into ~/.openclaw/workspace (SOUL.md and SKILL.md), producing persistent content in the user's OpenClaw workspace. This is a legitimate behavior for a skill but should be disclosed and confirmed before installation.
如何使用
  1. 确保已安装 OpenClaw(本地或 Docker 部署)
  2. 在对话框中输入安装命令:/install emp
  3. 安装完成后,直接呼叫该 Skill 的名称或使用 /emp 触发
  4. 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
v0.1.1
emp v0.1.1 - Updated homepage URL and metadata to point to a new repository and shell deploy method. - Added a shell-based “Deploy EMP Soul” step in install instructions for improved deployment. - No code or skill logic changes; documentation and metadata improvements only.
v0.1.0
EMP is a dual-purpose skill for the OpenClaw ecosystem. It stands for both **Employee** (dynamic task routing) and **Empathy** (Nonviolent Communication framework). It allows users to interact with a specialized "workforce" while ensuring all technical and interpersonal communication is grounded in **Nonviolent Communication (NVC)** principles.
元数据
Slug emp
版本 0.1.1
许可证
累计安装 0
当前安装数 0
历史版本数 2
常见问题

EMP 是什么?

EMP Skill for OpenClaw – empathetic, role-based task routing. 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 280 次。

如何安装 EMP?

在 OpenClaw 或 Claude Code 对话框中运行命令「/install emp」即可一键安装,无需额外配置。

EMP 是免费的吗?

是的,EMP 完全免费(开源免费),可自由下载、安装和使用。

EMP 支持哪些平台?

EMP 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 EMP?

由 Zingg(@zzj0402)开发并维护,当前版本 v0.1.1。

💬 留言讨论