Phone Controller | 手机操控者

WarnAudited by ClawScan on May 10, 2026.

Overview

The skill’s phone-control purpose is clear, but it asks users to install unpinned external Android-control code/APK and grant broad phone access.

Install only if you are comfortable granting broad Android ADB control and sending screenshots to the GLM provider. Pin and verify the external project/APK, use a test device if possible, avoid sensitive apps, confirm every write action carefully, and clean up USB debugging and keyboard settings afterward.

Findings (5)

Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.

What this means

A change or compromise in the external repository, dependencies, or APK could affect what code controls the phone and sees screen content.

Why it was flagged

The default setup installs unpinned external code and an external APK that are not included in the reviewed artifacts, yet those components will control an Android phone through ADB.

Skill content
git clone https://github.com/zai-org/Open-AutoGLM.git ... .venv/bin/pip install -r requirements.txt ... 下载 ADBKeyboard.apk:https://github.com/senzhk/ADBKeyBoard/blob/master/ADBKeyboard.apk
Recommendation

Only install from a pinned, reviewed release or commit; verify APK checksums/signatures; preferably use a test phone rather than a primary device.

What this means

If the agent misunderstands an instruction or the user confirms too quickly, it could affect apps, accounts, purchases, messages, or phone settings.

Why it was flagged

The skill can perform high-impact phone actions such as sending messages, ordering, paying, changing settings, and deleting data, but it documents a confirmation requirement before those actions.

Skill content
写入/发送操作(发消息、下单、支付、改设置、删除数据)→ 必须先向用户确认操作内容,获得明确同意后才执行
Recommendation

Use it for narrowly scoped tasks, review every proposed write/send/order/delete action, and avoid financial, password, and private-chat contexts.

What this means

Anyone who obtains the key may be able to use the user’s GLM account quota or access associated provider permissions.

Why it was flagged

The skill requires a GLM API key for the external model provider; this is expected for the stated integration but is still a sensitive credential.

Skill content
从 https://open.bigmodel.cn 获取 API Key,然后在命令中使用: ... --apikey "你的API_KEY"
Recommendation

Use a restricted API key if available, avoid sharing command history or logs containing the key, and rotate the key if exposure is suspected.

What this means

Sensitive information displayed on the phone could be transmitted to the provider during normal use.

Why it was flagged

The skill explicitly sends phone screenshots to an external model provider, which may include private app content, notifications, or account information visible on screen.

Skill content
手机屏幕截图会发送到外部 GLM API(智谱 https://open.bigmodel.cn)进行理解
Recommendation

Do not use it on banking, payment, password, medical, work-confidential, or private-chat screens; check the provider’s data policy before use.

What this means

If cleanup is skipped, the computer may retain debugging authorization and the ADB keyboard may remain active beyond the intended session.

Why it was flagged

USB debugging trust and the default ADB Keyboard setting can persist after a task; the skill discloses this and provides cleanup steps.

Skill content
点击「允许并始终允许」 ... 设为默认输入法 ... 使用后清理建议:撤销 USB 调试授权;关闭 USB 调试;切换回原输入法
Recommendation

After use, revoke USB debugging authorization, disable USB debugging, restore the original keyboard, and uninstall ADB Keyboard if it is no longer needed.