Openclaw Safety Guard

OpenClaw 安全卫士（发布 slug：openclaw-safety-guard，内部代号：openclaw-watchdog）：默认工作日 10:00 自动扫描 7 个维度（安全、记忆、心跳、定时任务、共享文件、通讯、代码规范），生成健康评分和可视化 Dashboard，通过飞书私信发送报告；没有 Cur...

Security Scan

The skill's files, runtime instructions, and required env/binaries are consistent with a local OpenClaw health-checker that posts reports to Feishu, but it performs broad local scans and will write cron/config state and run automatic low-risk fixes — review the setup and fix scripts before installing.

Purpose & Capability

The name/description promise a local watchdog that scans 7 dimensions and notifies via Feishu. Requested binaries (python3, node, npm) and environment variables (FEISHU_APP_ID, FEISHU_APP_SECRET) match that purpose (Python scripts for probes and a Node frontend build; Feishu credentials for notification/upload). The included scripts (scan_*.py, generate_dashboard.py, notify_feishu.py, upload_to_feishu_drive.py, setup.py) align with the declared functionality.

Instruction Scope

SKILL.md tells the agent to run setup.py (which writes config.json and registers a cron job), then run run_pipeline.py to execute probes that scan code, plist/launch agents, exec-approvals, git repos, memory/knowledge dirs and other local paths. Those reads and writes are within the stated watchdog scope, but they are broad (system LaunchAgents, exec-approvals file, workspace files). The agent will also obtain the installer’s Feishu open_id from the conversation context to populate notify.receive_id — this is expected for automated notifications but is an access-to-conversation-context action worth noting.

Install Mechanism

There is no external install spec (no arbitrary download step); code is provided in the skill bundle and SKILL.md instructs running local Python/Node commands. This is lower risk than fetching executables from untrusted URLs. No installers or third-party URL downloads are referenced in SKILL.md.

Credentials

Only FEISHU_APP_ID and FEISHU_APP_SECRET are declared and these are justified by the Feishu notification/drive upload functionality. However, the skill will read many local files/paths (plist, exec-approvals, workspace files, git repos) as part of scans; while coherent with the purpose, these are sensitive sources of data. The primaryEnv (FEISHU_APP_ID) is appropriate.

Persistence & Privilege

The skill’s post-install setup writes config.json and registers a cron job under .openclaw/state/cron/jobs.json and instructs restarting the Gateway so daily scans run. It also includes an automatic 'fix_green.py' remediation step (documented as limited to chmod-like low-risk ops). These are reasonable for a scheduled watchdog but constitute persistent presence and the ability to modify local state/permissions, so audit the setup and fix scripts before granting runtime access.

Scan Findings in Context

[base64-block] unexpected: A base64-block pattern was flagged in SKILL.md by the pre-scan. The visible SKILL.md content does not obviously contain base64 payloads, so this could be either a false positive from the scanner or an obfuscated/encoded block elsewhere in the documentation files. Treat this as a caution: inspect SKILL.md and any embedded strings in scripts (especially setup.py, notify_feishu.py, upload_to_feishu_drive.py, and fix_green.py) for encoded payloads or hidden instructions before installing.

Assessment

This skill is coherent with its stated purpose (local watchdog that notifies via Feishu) but it will: (1) read many local files (LaunchAgents plist, exec-approvals, git working tree, workspace knowledge directories), (2) write config.json and register a scheduled cron job, and (3) may perform automated low-risk fixes (chmod). Before installing: review the actual contents of scripts/setup.py, scripts/fix_green.py, scripts/notify_feishu.py and scripts/upload_to_feishu_drive.py to confirm they only access/ship data you expect and that fixes are limited to safe operations. Consider running the pipeline manually in a safe environment (python3 scripts/run_pipeline.py) with FEISHU env vars unset to inspect outputs first. Because the source and homepage are absent and a base64-block was flagged by the scanner, exercise caution: prefer manual code review or running the code in an isolated/test environment before granting it access to your real Feishu credentials and production workspace.

OpenClaw 安全卫士（openclaw-safety-guard）

OpenClaw 安全卫士的 Clawhub 发布 slug 为 openclaw-safety-guard，内部技术代号仍为 openclaw-watchdog。它是 OpenClaw 项目全链路健康监控工具，默认工作日自动扫描 7 个维度，生成评分和可视化 Dashboard，通过飞书发送报告。没有 Cursor 也可以直接查看 dashboard 和下载 JSON 日志。

安装（clawhub）

clawhub install openclaw-safety-guard

安装完成后，Agent 立即执行 post-install 初始化：

python3 {baseDir}/scripts/setup.py --receive_id <当前对话 sender 的 Feishu open_id>

--receive_id 由 Agent 从当前会话 context 中自动获取（即安装者本人的 open_id），无需手动填写。

setup.py 自动完成：

1. 写入 {baseDir}/config.json（绑定通知接收人）
2. 在 .openclaw/state/cron/jobs.json 注册每日扫描任务
3. 立即执行一次完整扫描，安装者 30 秒内收到第一份飞书健康报告

⚠️ 安装完成后重启 Gateway 使 cron 生效：openclaw gateway restart

前置条件

• FEISHU_APP_ID 和 FEISHU_APP_SECRET 已配置在 Gateway plist 的 EnvironmentVariables 中
• 本机已安装 node（用于首次 build frontend）
• OpenClaw Gateway 已运行

触发方式

1. 每日自动：安装时注册的 cron job，默认工作日 10:00（Asia/Shanghai）触发
2. 手动触发：在飞书群对 Agent 说 体检、生成健康大盘、跑一遍安全检查

执行流程

触发
  ↓
Step 1: 执行 7 个探针扫描
  python3 {baseDir}/scripts/run_pipeline.py
  内部依次调用：
    scan_heartbeat.py   → 心跳监控维度
    scan_standards.py   → 代码规范维度
    scan_memory.py      → 记忆健康维度
    scan_cron.py        → 定时任务维度
    scan_shared.py      → 共享文件维度
    scan_comm.py        → 通讯配置维度
    scan_security.py    → 安全维度
  ↓
Step 2: 聚合评分（含与上次对比的 score_delta）
    aggregate_watchdog.py
  ↓
Step 3: 生成可视化 Dashboard HTML（含飞书 bot 头像）
    generate_dashboard.py
  ↓
Step 4: 按次归档到 data/logs/YYYY-MM-DD_HH-MM/
  ↓
Step 5: 飞书私信通知（含健康分、问题列表、Dashboard 本地路径）
    notify_feishu.py
  ↓
Step 6: 自动执行 GREEN 安全修复（仅 chmod 类低风险操作）
    fix_green.py

手动重发通知

python3 {baseDir}/scripts/notify_feishu.py

配置文件

config.json 由 setup.py 自动生成，参考 config.example.json 了解全部可配项。

关键字段：

字段	说明
notify.receive_id	接收飞书通知的 open_id（setup.py 自动填入）
notify.score_recovery_threshold	分数提升超过此值时用「恭喜」语气（默认 3）
memory.checks[cross_workspace_conflict].tracked_concepts	自定义需要跨 workspace 监控一致性的概念
security.checks[knowledge_freshness].target	知识库目录路径（可选，不填则跳过此检查）

关键文件

文件	说明
data/latest_status.json	本次扫描完整结果（含 score_delta）
data/dashboard.html	最新 Dashboard，浏览器本地打开
data/history.json	问题生命周期历史（first_seen / resolved_at）
data/logs/YYYY-MM-DD_HH-MM/	按次归档的原始扫描日志（最多保留 30 次）
config.json	运行时配置（由 setup.py 生成，不提交到 git）
config.example.json	配置模板（可提交到 git）

错误处理

失败步骤	降级策略
探针执行报错	该维度显示 N/A，不影响其他维度
Dashboard 生成失败	通知仍发出，不含 Dashboard 路径
飞书通知失败	打印 ERROR 日志，pipeline 不中断
Bot info 获取失败	Dashboard 显示默认缩写，不影响报告发送