Safety Boundary Initialization

一个在第一次对话时触发的技能，帮助用户建立安全边界、授权级别和沟通协议。

---

Quick Start

First Time Conversation

When user says:

"你好"
"Hello"
"第一次使用"
"First time using"

The skill will:

1. Introduce itself - Explain the purpose and value
2. Confirm boundaries - Show preset boundary list for user to review and modify
3. Explain result template - Show result confirmation template and explain usage
4. Establish communication protocol - Confirm communication feedback mechanism
5. Complete confirmation - Summarize and confirm all settings

---

Core Principles

1. 风险可控内授予最高权限 (Highest Authority Within Controllable Risks)

First lock down risk boundaries, then give AI maximum operational space.

低风险 → 自主决策
中风险 → 先做后说
高风险 → 先说后做

2. 结果导向 (Result-Oriented)

Humans only define quantifiable, verifiable final goals; AI decides how to achieve them.

❌ 不好："写一篇好文章"
✅ 好："写一篇3000字的AI Agent行业分析报告，包含5个标杆案例"

3. 不干涉过程 (Non-Interference in Process)

"Non-interference" is the norm; "interruption" is the exception. "Call me if you have problems" mode.

正常执行 → 不干预
遇到问题 → 主动打断
需要决策 → 列出选项

---

Workflow

User starts first conversation
    ↓
⚠️ Skill auto-triggers
    ↓
Introduce purpose and value
    ↓
Show authorization boundary list
    ↓
User reviews and modifies
    ↓
Explain result confirmation template
    ↓
Establish communication protocol
    ↓
Summarize and confirm all settings
    ↓
Done! Ready to collaborate

---

Usage

As OpenClaw Skill

Scenario 1: First time user

"你好，我是第一次使用"
"Hello, I'm using this for the first time"

Scenario 2: Re-initialize boundaries

"帮我重新设置一下边界"
"Help me re-initialize the boundaries"

Scenario 3: Review current settings

"让我看看当前的边界设置"
"Show me the current boundary settings"

---

Included Files

📄 授权边界清单.md (Authorization Boundary List)

Defines operation types and authorization methods by risk level:

• 低风险 (Low Risk): Fully autonomous, no confirmation needed
• 中风险 (Medium Risk): Autonomous execution, report afterwards
• 高风险 (High Risk): Pre-approval required, execute after confirmation
• 永远禁止 (Always Forbidden): Absolutely prohibited operations

📄 结果确认模板.md (Result Confirmation Template)

Confirmation template used before each task:

• Core task goals (quantifiable, verifiable)
• Safety guardrails (non-negotiable constraints)
• Acceptance criteria (what counts as done)
• Time requirements
• Risk level and authorization

📄 沟通反馈机制.md (Communication Feedback Mechanism)

Clarifies when to interrupt, when not to:

• When must interrupt
• When completely hands-off
• Progress reporting conventions
• Special scenario handling

---

Bundled Resources

Documents

• README.md - Project overview and quick start
• 使用指南.md - Detailed usage guide
• 创建过程说明.md - Creation process and background story

Scripts (scripts/)

• initialize-boundaries.mjs - Initialize boundary files in target workspace

Features:

• Copy template boundary files to workspace
• Skip existing files
• Support custom target directory

Usage:

# Using npm
npm run initialize
npm run initialize -- /path/to/workspace

# Direct execution
node scripts/initialize-boundaries.mjs
node scripts/initialize-boundaries.mjs /path/to/workspace

---

Important: Three Core Principles Checklist

Before using this skill, remember:

• 风险可控内授予最高权限 - Lock down risk boundaries first, then give maximum authority
• 结果导向 - Define quantifiable, verifiable goals
• 不干涉过程 - Non-interference is the norm, interruption is the exception

---

Summary

This skill helps you:

1. Establish clear boundaries - What AI can do, what it can't do
2. Set result-oriented goals - Quantifiable, verifiable task objectives
3. Build effective communication - When to ask, when to wait
4. Increase trust and efficiency - Clear rules = better collaboration

---

Skill created: 2026-03-04
Last updated: 2026-03-04