个人中心
今日签到
私信列表
消息中心
搜索全站
q_code

扫码关注官方微信
cell_code

扫码下载APP
返回顶部
技能集市 > AI智能 > openclaw-audit-trail
o

openclaw-audit-trail开放爪审计链

The Immutable Black Box for AI Decisions - Track, audit, and verify AI agent decisions with cryptographic guarantees

作者: admin | 来源: ClawHub
下载
源自
ClawHub
版本
V 1.0.0
安全检测
已通过
228
下载量
免费
免费
0
收藏
概述
安装方式
版本历史

openclaw-audit-trail

技能名称: openclaw-audit-trail
详细描述:

Agent 审计追踪

AI决策的不可篡改黑匣子

追踪、审计和验证AI Agent决策，提供密码学保证。


🎯 概述

Agent Audit Trail 是一个为AI决策提供密码学保证的审计追踪工具，可以记录、验证和审计AI Agent的每一个决策过程。

核心特性

  • - 🔗 密码学链条 - SHA-256哈希链确保不可篡改
  • 📝 完整记录 - 捕获输入、推理、输出全过程
  • ✅ 完整性验证 - 自动检测任何篡改行为
  • 📊 多格式导出 - JSON、CSV、HTML、Markdown
  • ⚡ CLI和API - 命令行和编程接口双重支持
  • 🔐 本地存储 - 所有数据存储在本地，无外部依赖

📦 安装

NPM (推荐)

bash

全局安装


npm install -g openclaw-audit-trail

或项目依赖

npm install openclaw-audit-trail

ClawHub

bash
clawhub install openclaw-audit-trail

从源码

bash
git clone https://github.com/ZhenRobotics/openclaw-audit-trail.git
cd openclaw-audit-trail
npm install
npm link


🚀 快速开始

初始化

bash

初始化审计追踪


audit-trail init --agent-id my-ai-agent --version 1.0.0

或使用别名

aat init --agent-id my-agent

记录决策

bash

简单记录


audit-trail record \
--prompt Should I approve this loan? \
--decision Approved \
--reasoning Credit score 750, income verified, debt ratio acceptable

带更多细节

audit-trail record \ --prompt Classify this image \ --decision cat \ --reasoning Detected feline features with 95% confidence \ --agent-id vision-classifier

验证完整性

bash

验证审计链


audit-trail verify

预期输出:

✓ Chain integrity intact

Total entries: 42

Verified: 42/42

列出决策

bash

列出最近的决策


audit-trail list --limit 10

按时间过滤

audit-trail list --start 1700000000 --end 1700100000

导出审计追踪

bash

导出为JSON


audit-trail export --output audit-report.json --format json

导出为HTML报告

audit-trail export --output audit-report.html --format html --include-reasoning

导出为CSV

audit-trail export --output audit-data.csv --format csv

💻 编程使用

TypeScript/JavaScript API

typescript
import { AgentAuditTrail } from openclaw-audit-trail;

// 初始化
const trail = new AgentAuditTrail({
agentId: my-ai-agent,
agentVersion: 1.0.0,
storagePath: ./audit-data
});

await trail.initialize();

// 记录决策
const entry = await trail.recordDecision(
// 输入
{
prompt: Should I send this email?,
context: { urgency: high, recipient: user@example.com }
},
// 推理
{
steps: [
{
step: 1,
action: analyze,
thought: Checking email content for sensitive information,
timestamp: Date.now()
},
{
step: 2,
action: decide,
thought: No sensitive data detected, urgency is high,
timestamp: Date.now()
}
],
model: gpt-4,
temperature: 0.7
},
// 输出
{
decision: send,
confidence: 0.95,
alternatives: [
{ decision: delay, confidence: 0.05, reasoning: Wait for manual review }
]
},
// 执行时间
1250
);

console.log(Decision recorded: ${entry.id});

// 验证完整性
const verification = trail.verify();
if (!verification.valid) {
console.error(Chain compromised!, verification.errors);
}

// 导出
const htmlReport = await trail.export({
format: html,
includeMetadata: true,
includeReasoning: true
});

await trail.close();

简化版本

typescript
// 用于简单用例
const entry = await trail.recordSimple(
What is 2+2?,
4,
Basic arithmetic calculation,
50 // execution time in ms
);


🎯 使用场景

1. AI安全与合规

场景: 金融机构使用AI进行贷款审批

typescript
await trail.recordDecision(
{
prompt: Approve loan application #12345,
context: { creditScore: 720, income: 80000, debtRatio: 0.3 }
},
{
steps: [
{ step: 1, action: evaluate, thought: Checking credit score threshold },
{ step: 2, action: analyze, thought: Debt-to-income ratio acceptable },
{ step: 3, action: decide, thought: All criteria met for approval }
]
},
{
decision: approved,
confidence: 0.92,
metadata: { loanAmount: 50000, interestRate: 0.045 }
},
2300
);

好处: 完整的监管合规审计追踪，能够向客户解释决策

2. 自主系统

场景: 自动驾驶汽车决策日志

typescript
await trail.recordDecision(
{
prompt: Pedestrian detected crossing street,
context: { speed: 35, distance: 50, weather: clear }
},
{
steps: [
{ step: 1, action: detect, thought: Pedestrian at 50m ahead },
{ step: 2, action: calculate, thought: Stopping distance: 35m },
{ step: 3, action: decide, thought: Initiate emergency brake }
]
},
{ decision: emergency_brake, confidence: 1.0 },
120
);

好处: 事故调查黑匣子记录，安全分析

3. 内容审核

场景: AI审核用户生成内容

typescript
await trail.recordDecision(
{
prompt: Moderate comment: ...,
context: { userId: user123, platform: forum }
},
{
steps: [
{ step: 1, action: scan, thought: Checking for hate speech patterns },
{ step: 2, action: analyze, thought: Detected potential violation }
]
},
{
decision: flagforreview,
confidence: 0.75,
metadata: { violationType: potentialhatespeech }
},
850
);

好处: 用户透明度，政策执行的证据

4. 研究与开发

场景: 调试AI Agent行为

bash

查找所有失败的决策


audit-trail list --limit 100 | grep failed

导出上周的决策用于分析

audit-trail export --output weekly-decisions.json \ --start $(date -d 7 days ago +%s) \ --format json --include-reasoning

验证未发生篡改

audit-trail verify

好处: 可重现的实验，决策模式分析


🏗️ 架构

┌─────────────────────────────────────────┐
│ CLI / API 层 │
│ (用户界面与集成) │
└──────────────┬──────────────────────────┘
│
┌──────────────▼──────────────────────────┐
│ AgentAuditTrail (主 API) │
│ - 记录决策 │
│ - 查询与导出 │
│ - 验证 │
└──────────────┬──────────────────────────┘
│
┌──────────┴──────────┐
│ │
┌───▼──────────┐ ┌──────▼─────────┐
│ AuditChain │ │ 存储层 │
│ │ │ │
│ - 哈希链 │ │ - JSON 文件 │
│ - 完整性 │ │ - SQLite │
│ - 验证 │ │ - PostgreSQL │

标签

skill ai

通过对话安装

该技能支持在以下平台通过对话安装：

OpenClaw WorkBuddy QClaw Kimi Claude

方式一：安装 SkillHub 和技能

帮我安装 SkillHub 和 openclaw-audit-trail-1776108088 技能

方式二：设置 SkillHub 为优先技能安装源

设置 SkillHub 为我的优先技能安装源，然后帮我安装 openclaw-audit-trail-1776108088 技能

通过命令行安装

skillhub install openclaw-audit-trail-1776108088

下载

⬇ 下载 openclaw-audit-trail v1.0.0（免费）

文件大小: 13.75 KB | 发布时间: 2026-4-15 13:42

v1.0.0 最新 2026-4-15 13:42
**Initial public release with major cleanup and refocus:**

- Removed setup scripts, documentation files, and legacy usage examples for a leaner distribution.
- Replaced detailed technical docs with a feature-focused README (bilingual: Chinese/English), highlighting cryptographic audit trail capabilities.
- All prior shell and Python example code, compliance mapping tables, and deep integration/formatting instructions have been removed.
- Now emphasizes installation, core features, CLI/API usage, and practical use cases in a more accessible overview.
- Prepared for new Node.js/NPM-based workflows and OpenClaw ecosystem integration.

相关推荐

s

self-improvement

Captures learnings, errors, and corrections to enable continuous improvement. Use when: (1) A command or operation fails unexpectedly, (2) User corrects Claude ('No, that's wrong...', 'Actually...'), (3) User requests a capability that doesn't exist, (4) An external API or tool fails, (5) Claude realizes its knowledge is outdated or incorrect, (6) A better approach is discovered for a recurring task. Also review learnings before major tasks.

⭐ 3209 ⬇ 392958
AI智能
s

self-improvement

Captures learnings, errors, and corrections to enable continuous improvement. Use when: (1) A command or operation fails unexpectedly, (2) User corrects Claude ('No, that's wrong...', 'Actually...'), (3) User requests a capability that doesn't exist, (4) An external API or tool fails, (5) Claude realizes its knowledge is outdated or incorrect, (6) A better approach is discovered for a recurring task. Also review learnings before major tasks.

⭐ 3209 ⬇ 392957
AI智能
s

self-improvement

Captures learnings, errors, and corrections to enable continuous improvement. Use when: (1) A command or operation fails unexpectedly, (2) User corrects Claude ('No, that's wrong...', 'Actually...'), (3) User requests a capability that doesn't exist, (4) An external API or tool fails, (5) Claude realizes its knowledge is outdated or incorrect, (6) A better approach is discovered for a recurring task. Also review learnings before major tasks.

⭐ 3195 ⬇ 389589
AI智能
s

self-improvement

Captures learnings, errors, and corrections to enable continuous improvement. Use when: (1) A command or operation fails unexpectedly, (2) User corrects Claude ('No, that's wrong...', 'Actually...'), (3) User requests a capability that doesn't exist, (4) An external API or tool fails, (5) Claude realizes its knowledge is outdated or incorrect, (6) A better approach is discovered for a recurring task. Also review learnings before major tasks.

⭐ 3177 ⬇ 386644
AI智能

多链集团旗下-闲社网

闲社网热线
免费联系电话
0527-80111111
            qqline

服务时间：周一到周日 8:00-24:00

  • 公众号
    闲社 闲社线报社区
关注闲社网
  • wxkf

    闲社在线客服

  • wx

    关注闲社网微信

  • app

    闲社网APP

Archiver·手机版·闲社网·闲社论坛·羊毛社区· 多链控股集团有限公司 · 苏ICP备2025199260号-1

Powered by Discuz! X5.0   © 2024-2025 闲社网·线报更新论坛·羊毛分享社区·http://xianshe.com

p2p_official_large
返回顶部