SafeHub

SafeHub 是面向 OpenClaw Skill 的安全扫描器：对技能名称、本地路径或 GitHub 仓库运行 Semgrep 静态分析，并在 Docker 可用时执行沙箱行为观察，汇总网络、文件系统、eval/exec、环境变量与混淆类发现，给出 0-100 信任分及安装建议；另支持回看历史报告和更新规则。

sumeetghimire

热度 56.89

最近更新 2026-05-18 12:17:31

分类：安全依赖：代码执行 / 读写文件 / 第三方 API应用场景：代码审查 / DevOps / 自动化脚本 / 报告撰写

安装技能

复制下面的话术发给你的 Agent，它会通过 secure-skills 检索并安装该技能。

执行命令安装此 skill： npx secure-skills-cli add safehub

Metadata

name: safehub
description: Scan OpenClaw skills for malware and security issues before installation. Use when the user wants to verify a skill is safe, audit a ClawHub skill, or check a local or GitHub skill.
version: 1.0.2
author: sumeetghimire
metadata: {"openclaw":{"emoji":"🛡️","requires":{"bins":["node","semgrep","git"]},"os":["darwin","linux","win32"]}}

SafeHub

SafeHub is a security scanner for OpenClaw skills. It runs static analysis (Semgrep) and optional sandbox execution (Docker) on any skill—by name, local path, or GitHub URL—and returns a trust score and a clear recommendation: safe to install, install with caution, or not safe.

Requirements

These binaries must be on your PATH (declared in registry metadata):

Node.js (18+) — required to run the CLI.
Semgrep — required for the scan command (static analysis). Install with brew install semgrep or npm install -g semgrep.
git — required when the scan target is a GitHub URL (used to clone the repo).

Optional:

Docker — used for sandbox execution. If Docker is not available, use --no-sandbox for static-only scanning.

Environment variables

All of these are optional. No secrets or API tokens are required by default.

Variable	Default	Effect
SAFEHUB_RULES_REPO	`safehub/safehub`	GitHub repo (owner/repo) used by `safehub update` to fetch and overwrite local rule files in `./rules`. Setting this to another repo makes the updater pull rules from that repo — use only repos you trust.
SAFEHUB_RULES_BRANCH	`main`	Branch name used when fetching rules (with `SAFEHUB_RULES_REPO`).
SAFEHUB_DATA_DIR	`~/.safehub`	Directory for cached scan reports (e.g. `~/.safehub/reports`).
SAFEHUB_SANDBOX_IMAGE	`node:18-alpine`	Docker image used for the sandbox when scanning.
SAFEHUB_SANDBOX_TIMEOUT_MS	`30000`	Timeout (ms) for the sandbox run before the container is killed.
SAFEHUB_NO_TYPING	(unset)	Set to `1` to disable the typing-effect output (e.g. in CI or pipes).

Important: SAFEHUB_RULES_REPO controls where safehub update downloads rules from and overwrites local ./rules; only point it at a repo you trust.

Commands

All commands are run via the safehub CLI (e.g. safehub scan <target> or node index.js scan <target> from the skill directory).

scan

Scan a skill by ClawHub name, local path, or GitHub URL.

Examples:

safehub scan web-scraper
safehub scan ./my-local-skill
safehub scan https://github.com/user/their-skill
safehub scan https://github.com/BenedictKing/tavily-web --no-sandbox

Options:

--no-sandbox — Skip Docker sandbox; run static analysis only (use when Docker is not installed).

report

Show the last scan report for a skill without rescanning.

Examples:

safehub report web-scraper
safehub report risky-skill

update

Pull the latest Semgrep scanner rules from the SafeHub GitHub repo (or your fork via SAFEHUB_RULES_REPO).

Examples:

safehub update
SAFEHUB_RULES_REPO=owner/repo safehub update

Example output

After running safehub scan <target>, you’ll see:

Static analysis — Findings from Semgrep (network, filesystem, eval/exec, env, obfuscation).
Sandbox behavior — Whether the skill attempted network access or suspicious actions (when Docker is used).
Trust score (0–100) and recommendation: SAFE TO INSTALL, INSTALL WITH CAUTION, or NOT SAFE TO INSTALL.

Installation (users)

Install from ClawHub:

clawhub install safehub

Or install the CLI globally from npm:

npm install -g safehub

Then run safehub scan <target> (if the CLI is on PATH) or node index.js scan <target> from the skill directory.