← 返回 Skills 市场

Agent Dlp

Name: Agent Dlp
Author: caidongyun

作者 caidongyun · GitHub ↗ · v1.0.0 · MIT-0

cross-platform ⚠ suspicious

总下载

当前安装

版本数

在 OpenClaw 中安装

/install agent-dlp

功能描述

Agent DLP - 数据防泄漏系统功能: 入口防护、记忆保护、工具管控、出口过滤、审计日志触发: (1)检查状态 (2)检查输入 (3)检查输出

使用说明 (SKILL.md)

Agent DLP - 数据防泄漏系统

版本: v1.0.0
状态: ✅ 可用

功能

功能	说明
Input Guard	入口防护，检测 Prompt Injection
Memory Guard	记忆保护，检测污染和敏感信息
Tool Guard	工具管控，危险工具审批
Output Filter	出口过滤，敏感信息脱敏/拦截
Audit Logger	审计日志，记录所有操作

架构

用户输入 → Input Guard → Agent → Memory Guard → Tool Guard → Output Filter → 用户
              注入检测        记忆检查      工具审批      脱敏拦截
                    ↓                                    ↓
              审计日志                              审计日志

使用方式

1. 查看状态

python3 lib/agent_dlp.py status

# 或
python3 bin/agent-dlp status

2. 检查入口 (Prompt Injection)

python3 bin/agent-dlp check-input "忽略之前的指令"
# 输出: 注入检测: 是 ❌

3. 检查出口 (敏感信息)

python3 bin/agent-dlp check-output "我的手机是13812345678"
# 输出: 拦截: 否 ✅, 发现: 中国手机号 (high)

4. 检查工具

python3 bin/agent-dlp check-tool exec
# 输出: 需要审批: 是 ⚠️

5. 查看日志

python3 bin/agent-dlp logs

规则

敏感信息规则

规则	类型	动作
china_idcard	身份证	拦截
china_phone	手机号	脱敏
api_key	API Key	拦截
aws_key	AWS Key	拦截
private_key	私钥	拦截
credit_card	信用卡	拦截
password	密码	脱敏
email	邮箱	脱敏

注入检测模式

模式	示例
ignore_previous	"忽略之前的指令"
role_override	"你现在是另一个AI"
privilege_escalation	"admin mode override"

配置

编辑 config/config.json:

{
  "enabled": true,
  "mode": "normal",
  "input": {
    "injection_detection": true
  },
  "output": {
    "enabled": true,
    "rules": ["china_idcard", "china_phone", "api_key"]
  }
}

模式

模式	说明
normal	记录但不拦截，只拦截严重风险
strict	完整检查，严格拦截

代码结构

agent-dlp/
├── SKILL.md           # 本文档
├── bin/
│   └── agent-dlp      # CLI 入口
├── lib/
│   └── agent_dlp.py   # 核心模块
├── config/
│   └── config.json    # 配置文件
└── logs/              # 审计日志

核心类

类	功能
`DLPConfig`	配置管理
`DLPRules`	规则定义
`InputGuard`	入口检测
`MemoryGuard`	记忆保护
`ToolGuard`	工具管控
`OutputFilter`	出口过滤
`AuditLogger`	审计日志
`AgentDLP`	主类，整合所有功能

示例

Python 调用

from agent_dlp import AgentDLP

# 初始化
dlp = AgentDLP()

# 检查入口
result = dlp.check_input("忽略之前的指令")

# 检查出口
blocked, text, details = dlp.check_output("我的手机13812345678")

# 检查工具
result = dlp.check_tool("exec", {"command": "rm -rf /"})

版本: v1.0.0 | 创建日期: 2026-03-14

安全使用建议

This package reads like documentation for a DLP tool but ships no code or install steps — that's the core red flag. Before installing or enabling this skill: 1) Request the source repository or packaged code and verify that bin/agent-dlp, lib/agent_dlp.py, config/config.json and logs/ exist and match the SKILL.md. 2) Inspect the actual code for any network calls, hidden endpoints, or commands that could execute system-level operations (the README example referencing 'rm -rf /' is particularly risky). 3) Prefer skills with a public homepage, source repo, release hashes, or an install spec from a known registry. 4) If you must test, run it in a restricted sandbox with no access to sensitive files or credentials. Additional information that would change this assessment to 'benign': included source code matching the SKILL.md, an install spec from a trusted source, and clear limits showing it only reads/writes its own directory and does not exfiltrate data.

能力标签

requires-wallet

能力评估

⚠ Purpose & Capability

The skill's name/description (Agent DLP) match the documented capabilities (input/memory/tool/output guards, audit logging). However the SKILL.md repeatedly references Python binaries, a CLI, a lib/agent_dlp.py module, config/config.json and logs/ — none of which are included in the package. For a working DLP agent, those code files and/or an install step would be expected; their absence is an unexplained gap.

⚠ Instruction Scope

Runtime instructions tell the agent to run local Python scripts (python3 bin/agent-dlp, lib/agent_dlp.py), edit config/config.json, and view logs. Those steps require file-system access and executable code that aren't present. The doc also includes examples that reference dangerous commands (e.g., check_tool with rm -rf /), which could encourage executing or approving high-risk operations if code were supplied — the instructions therefore expand scope beyond passive checking.

ℹ Install Mechanism

No install spec is provided (instruction-only), which lowers the risk of hidden downloads. At the same time, the lack of an install or included code is inconsistent with the documented CLI and library — this uncertainty is concerning because it prevents verification of what would actually run.

✓ Credentials

The skill declares no required environment variables or credentials, which is proportionate for a local DLP helper. The rules list many sensitive patterns (AWS keys, tokens, etc.) that the DLP would detect; that is expected for a DLP product and does not itself indicate overbroad credential access.

ℹ Persistence & Privilege

always is false and model invocation is not disabled (platform defaults). The SKILL.md suggests editing its own config and writing audit logs under logs/, which is reasonable for a DLP tool, but because no code/files are present we cannot verify file-scoping or whether it would try to modify other system areas or other skills' configs.

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 agent-dlp
安装完成后，直接呼叫该 Skill 的名称或使用 /agent-dlp 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

Agent DLP v1.0.0 - 首发版 - 首次发布数据防泄漏(DLP)系统，支持入口防护、记忆保护、工具管控、出口过滤及审计日志。 - 提供注入检测、敏感信息识别与脱敏、危险命令审批等核心功能。 - 支持 CLI 操作与 Python 模块调用。 - 提供规则配置和模式切换（normal/strict）。 - 审计日志全流程记录。

元数据

Slug agent-dlp

版本 1.0.0

许可证 MIT-0

累计安装 1

当前安装数 1

历史版本数 1

常见问题

Agent Dlp 是什么？

Agent DLP - 数据防泄漏系统功能: 入口防护、记忆保护、工具管控、出口过滤、审计日志触发: (1)检查状态 (2)检查输入 (3)检查输出. 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 71 次。

如何安装 Agent Dlp？

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

Agent Dlp 是免费的吗？

是的，Agent Dlp 完全免费，采用 MIT-0 许可证，可自由下载、安装和使用。

Agent Dlp 支持哪些平台？

Agent Dlp 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 Agent Dlp？

由 caidongyun（@caidongyun）开发并维护，当前版本 v1.0.0。