OpenClaw Security Handbook (CN)

基于 ZAST.AI 安全手册的 OpenClaw 安全审计与加固技能。运行全面安全诊断（内置 audit + 手册补充项），生成结构化报告，提供交互式修复引导，支持定时审计调度。触发场景：安全审计、安全加固、漏洞检查、security audit、hardening、暴露检查。

Security Scan

The skill's requests and instructions are coherent with a local OpenClaw security-audit / hardening tool; it reads and inspects local OpenClaw state but does not request unrelated credentials or contact external endpoints by default.

Purpose & Capability

Name/description match the actual behavior: it runs `openclaw security audit --json`, parses results, performs local checks under ~/.openclaw, and offers remediation guidance. The files included (report generator and checklist/fix docs) are consistent with a security handbook/diagnostic skill.

Instruction Scope

The SKILL.md and scripts instruct the agent to read many local files (configs, sessions, logs, memory, installed skills), run local network/port checks, and inspect a sandbox docker container. This is expected for a security audit, but these actions will access potentially sensitive data (API keys, session logs, MEMORY.md). The skill emphasizes user confirmation before destructive changes, which is appropriate.

Install Mechanism

No install spec; instruction-only skill with two helper scripts. Nothing is downloaded or written by an automated installer. The publish script calls packaging/publishing tools on-demand and prompts for confirmation.

Credentials

The skill declares no required env vars or credentials, and its checks operate on local OpenClaw files only (proportional). One caution: the provided fix examples include commands that append tokens to ~/.openclaw/.env (echo >>), which would store secrets in plaintext if used; the docs also advise not to write plaintext, a slight contradiction that the user should review before running.

Persistence & Privilege

always:false and no system-wide modifications by default. Some recommended manual remediation steps (chattr, iptables, rm -rf, revoking keys) require elevated privileges or are destructive, but SKILL.md states these need explicit user confirmation.

Assessment

This skill is internally consistent for performing a local OpenClaw security audit, but it will read sensitive local files (logs, session files, MEMORY.md, installed skill sources) to look for secrets and risky code. Before running: 1) ensure you trust the skill and run it on a machine where exposing those files to the agent is acceptable (prefer a disposable VM); 2) review the included scripts and the commands in references/fix-commands.md — especially any commands that write tokens to ~/.openclaw/.env or delete directories; 3) do not blindly accept automatic fixes for operations that modify tokens/configs or remove files — back up before applying fixes; 4) the publish script will attempt to package/publish only if you run it and confirm. If you want minimal risk, run the audit commands manually and inspect the generated JSON/report rather than allowing automated fix actions.

OpenClaw Security — 安全审计与加固

概述

运行全面安全诊断，生成结构化报告，提供交互式修复引导。核心信号来自 openclaw security audit --json，补充手册（ZAST.AI 安全研究团队）中的手动检查项。

核心原则：

• 所有修复操作需用户确认，不自动执行
• 安全默认值修复（--fix）可安全推荐
• 密钥/凭证操作、Skill 卸载、配置文件删除等为不可逆操作，必须人工决策
• 优先使用 JSON 输出（--json）做机器解析

工作流

0. 开始前确认

用中文询问用户意图（列出选项，回复数字）：

请选择操作模式：
1. 完整审计（推荐）：诊断 + 报告 + 修复引导
2. 快速诊断：仅运行 openclaw security audit --deep
3. 交互式修复：基于上次报告逐项修复
4. 调度审计：配置定时安全检查

1. 运行诊断（并行执行）

按顺序执行以下检查，结果汇总后统一呈现：

A. OpenClaw 内置审计（JSON 格式）

openclaw security audit --json

解析 JSON，提取 findings[]（severity/checkId/title/detail/remediation）。

B. 补充手动检查项

以下检查项内置 --fix 不覆盖，需手动扫描：

B-1. 文件权限

# 检查关键文件和目录权限
ls -la ~/.openclaw/
stat -c "%a %n" ~/.openclaw/openclaw.json
stat -c "%a %n" ~/.openclaw/.env
stat -c "%a %n" ~/.openclaw/credentials/

B-2. 日志/会话密钥泄露扫描

grep -rEn "sk-|AKIA-|password|secret|token|private.key" ~/.openclaw/sessions/ 2>/dev/null | grep -v ".jsonl:" | head -20
grep -rEn "sk-|password|secret|token" ~/.openclaw/logs/ 2>/dev/null | head -20

B-3. 附件文件权限

find ~/.openclaw/ -type f \( -name "*.jpg" -o -name "*.png" -o -name "*.pdf" -o -name "*.doc*" \) -perm +044 2>/dev/null

B-4. 端口暴露检查（本地）

# 检查网关 + 沙盒浏览器端口
ss -ltnp 2>/dev/null | grep -E "18789|18790|9222|5900|6080" || netstat -tlnp 2>/dev/null | grep -E "18789|18790|9222|5900|6080"

B-5. 记忆文件完整性

cat ~/.openclaw/workspace/MEMORY.md 2>/dev/null | head -50
ls -lt ~/.openclaw/workspace/memory/ 2>/dev/null | head -10

B-6. Skill 危险模式扫描

# 检查可疑文件模式（低权限扫描，不执行）
find ~/.openclaw/skills/ -name "*.js" -o -name "*.ts" -o -name "*.py" 2>/dev/null | head -50

B-7. Docker 沙箱网络模式（如使用）

docker inspect openclaw-sandbox 2>/dev/null | grep -A5 NetworkMode || echo "无沙箱容器"

2. 生成报告

将所有发现分为三组，格式化输出：

🔴 CRITICAL（必须立即处理） 🟡 WARN（建议处理） ℹ️ INFO（供参考）

每项输出格式：

[checkId] 标题
  当前状态: <实际值>
  建议: <修复方法>
  修复命令: <具体命令或 "需人工确认">

3. 交互式修复引导

按严重程度逐项处理。每项询问（数字选择）：

修复项：[标题]
当前: <当前值>
建议: <修复方法>
修复命令: <具体命令>

选择：
1. 执行此修复
2. 跳过
3. 详细信息

可安全自动化的项目（直接推荐 1）：

• chmod 600/700 文件权限
• openssl rand -hex 32 生成强令牌
• DISABLE_TELEMETRY=1 环境变量
• 日志密钥泄露（确认后 grep 删除）

必须人工确认的项目（默认 2 跳过）：

• groupPolicy 修改
• sandbox.mode 变更
• Skill 卸载/安装
• OAuth token revoke
• 清除 workspace/sessions

4. openclaw security audit --fix

推荐执行内置安全默认值修复：

openclaw security audit --fix

说明：--fix 只修复文件权限和配置默认值，不执行破坏性操作。

5. 调度审计（可选）

完成后询问（数字选择）：

是否需要调度定期安全审计？
1. 每天 09:00 自动审计（openclaw cron）
2. 每周一 09:00 审计
3. 不调度

如选择调度：

openclaw cron add --name "openclaw-security:daily" --command "openclaw security audit --json" --cron "0 9 * * *" --output ~/.openclaw/logs/security-audit-$(date +\%Y\%m\%d).json

安全检查清单（完整版）

详细检查项见 references/checklist.md。

修复命令参考

各类问题的修复命令见 references/fix-commands.md。

脚本说明

• scripts/security-report.py — 解析 --json 输出并格式化，支持 --fix 模式自动应用安全默认值
• references/checklist.md — 完整安全检查清单（手册第九章）
• references/fix-commands.md — 各类修复命令速查表

关于本 Skill

基于 ZAST.AI OpenClaw Security Handbook 构建，集成 OpenClaw 内置 security audit 命令，补充手册中的手动检查项。