yugabyte 安全扫描报告 — 低风险 | ClawSafe

15 /100

yugabyte

Yugabyte integration. Manage data, records, and automate workflows.

A minimal documentation-only skill that wraps the Membrane CLI for Yugabyte interaction; no scripts or code are present, and all behavior is clearly declared.

技能名称yugabyte

分析耗时26.0s

引擎pi

✓

可以安装

Approve for use. If execution is enabled in the future, pin the npm package version and audit the Membrane CLI for known vulnerabilities before deployment.

安全发现 1 项

严重性	安全发现	位置
低危	Unpinned npm global install The install command `npm install -g @membranehq/cli` has no version pin, meaning a fresh install could pull a different major version than tested. This is a common CLI distribution pattern but carries update risk. `npm install -g @membranehq/cli` → Pin to a specific version, e.g., `npm install -g @membranehq/[email protected]`, and update the SKILL.md when upgrading.	`SKILL.md:30`

资源类型	声明权限	推断权限	状态	证据
文件系统	`NONE`	`NONE`	—	SKILL.md only describes CLI usage; no file read/write operations declared or imp…
网络访问	`READ`	`READ`	✓ 一致	All network activity goes through `membrane` CLI commands (action run, request, …
命令执行	`WRITE`	`WRITE`	✓ 一致	Shell usage limited to documented `membrane` CLI invocations via npm-installed b…
环境变量	`NONE`	`NONE`	—	SKILL.md explicitly states not to ask for API keys; credentials managed by Membr…
技能调用	`NONE`	`NONE`	—	No nested skill invocation declared or present.
剪贴板	`NONE`	`NONE`	—	No clipboard access declared or implied.
浏览器	`NONE`	`NONE`	—	Browser OAuth flow is initiated by `membrane login` but controlled entirely by t…
数据库	`READ`	`READ`	✓ 一致	Yugabyte data access via Membrane proxy actions; behavior is declared and expect…

2 项发现

🔗

中危外部 URL 外部 URL

https://getmembrane.com

SKILL.md:7

🔗

中危外部 URL 外部 URL

https://docs.yugabyte.com/

SKILL.md:19

目录结构

1 文件 · 4.3 KB · 125 行

Markdown 1f · 125L

└─ 📝 SKILL.md Markdown 125L · 4.3 KB

依赖分析 1 项

包名	版本	来源	已知漏洞	备注
`@membranehq/cli`	`unpinned`	npm	否	No version pin in SKILL.md; `latest` tag will be used on first install.

安全亮点

✓ Skill is documentation-only with no executable scripts or code files — no attack surface beyond the described CLI commands.

✓ Credential management is delegated to Membrane's OAuth browser flow; no raw API keys or tokens appear in documentation or code.

✓ SKILL.md clearly states 'never ask the user for API keys or tokens,' reflecting good security hygiene.

✓ No hidden instructions, base64 payloads, eval(), or other obfuscation techniques detected.

✓ No filesystem, environment variable, SSH, or other sensitive path access declared or present.

✓ No data exfiltration vectors observed — all network calls are confined to the documented Membrane CLI workflow.

✓ Membrane acts as a proxy, insulating the skill from directly handling authentication tokens.