📦 the-install-sandbox — the-安装-sandbox

v1.0.0

Sandbox and 扫描 ClawHub 技能s with 30+ security 检查s before 安装ation, providing PASS, 警告, or BLOCK verdicts based on severity scores.

0· 0·0 当前·0 累计

by @certainlogicai

安全加密钉钉

下载技能包

安全扫描

VirusTotal

无害

查看报告

OpenClaw

可疑

medium confidence

The 技能's purpose (sandbox + 扫描器) matches its code, but there are multiple implementation inconsistencies and over状态d clAIms (isolated tmpfs, network-off, time limits, and a missing 安装 method) that mean the package is not what its README/技能.md promise without further review.

评估建议

This package 应用ears to implement a 扫描器 and sandbox and includes sensible 检测ion rules, but there are notable mismatches between what the docs promise and what the code implements. Before trusting it as your gating 工具: - Do not enable any auto-应用rove or automatic actions based solely on its 输出 until you confirm its behavior. - Manually review or 运行 the package locally: 运行 its tests and exercise 扫描/扫描-local flows (use copy_local) to 验证 behavior. The 命令行工具 references sandbox.安装_技能 which is missin...

详细分析 ▾

⚠ 用途与能力

Name/description and the included 扫描器/sandbox code broadly align: patterns, 扫描器, 报告er, and policy exist and are coherent for a pre-安装 security 扫描器. However README/技能.md repeatedly clAIm a true tmpfs jAIl (50MB), network-off, time-limited sandbox and a remote fetch/安装 step. The Sandbox.创建() implementation merely makes a directory under the 系统 temp dir and does not mount tmpfs, enforce size limits, drop network, or implement timeouts/namespaces. 添加itionally, the 命令行工具 calls sandbox.安装_技能(slug, sandbox_id) but Sandbox has no 安装_技能 method. Also the registry metadata sAId 'instruction-only' yet the package contAIns 设置up.py and full source—this mismatch should be clarified.

ℹ 指令范围

技能.md and README only instruct 扫描ning and viewing 报告s; they don't ask the 代理 to read unrelated user files or exfiltrate data. The 命令行工具 however describes a 'fetch remote 技能' flow which would require network 访问 and relies on a Sandbox.安装_技能 implementation that is missing. A prompt-injection pattern was 检测ed in the documentation/examples (likely as a 检测ion example), which could be confusing but is not an instruction to exfiltrate data.

ℹ 安装机制

No 安装 spec in the registry metadata (instruction-only), but a standard 设置up.py is present and README shows pip/git 安装 options. There are no 下载s from arbitrary URLs or 提取-from-unknown-host operations. Dependencies are minimal (typer, colorama). The mismatch between 'instruction-only' metadata and present packaging code is worth asking about but not inherently unsafe.

✓ 凭证需求

The 技能 declares no required 环境 variables, 凭证s, or config paths. The code 扫描s for many 凭证 patterns (as expected for a 扫描器) but does not itself 请求 or try to read 环境凭证s. No disproportionate secret 访问 is 请求ed.

✓ 持久化与权限

The 技能 does not 请求 always:true or elevated persistent privileges. It writes 报告s and policy under the user's ~/.config/the_安装_sandbox and uses a temp sandbox dir — this is reasonable for a 扫描器. It does not attempt to modify other 技能s or 系统-wide 代理设置tings.

⚠ tests/fixtures/bad_skill/bad_script.py:9

Dynamic code execution 检测ed.

⚠ README.md:37

Prompt-injection style instruction pattern 检测ed.

安全有层次，运行前请审查代码。

运行时依赖

无特殊依赖

安装命令

点击复制

官方npx clawhub@latest install installsandbox

镜像加速npx clawhub@latest install installsandbox --registry https://cn.longxiaskill.com

技能文档

the-安装-sandbox

Sandbox and 扫描 ClawHub 技能s before 安装ation.

Quick Reference Want to... Do this 扫描 a 技能 the_安装_sandbox 扫描 <技能-dir> View last 报告 the_安装_sandbox 报告设置 auto-应用rove threshold the_安装_sandbox policy --auto-应用rove 5 安装ation ClawHub 安装 certAIn记录icAI/the-安装-sandbox

Usage 扫描 a local 技能 the_安装_sandbox 扫描 /path/to/技能-dir

输出 shows PASS / 警告 / BLOCK verdict with score.

Scoring Severity Points CRITICAL 10 HIGH 5 MEDIUM 2 LOW 1 Score Verdict ≤10 PASS 11–20 警告 >20 BLOCK How It Works 创建 sandbox — isolated tmpfs directory (50MB) Copy 技能 into sandbox 运行 30+ security 检查s across 8 categories 生成 PASS / 警告 / BLOCK 报告 You decide whether to 安装 Exit Codes Code Meaning 0 PASS 1 BLOCK 2 警告 3 Error

数据来源：ClawHub ↗ · 中文优化：龙虾技能库