Gemini Image Gen + Watermark Removal

ReviewAudited by ClawScan on May 10, 2026.

Overview

The main Gemini image workflow is disclosed, but the skill includes high-impact account/browser use and an optional hard-coded Feishu send step that users should review carefully.

Install only if you are comfortable letting the agent control a logged-in Google browser for Gemini. Verify any GeminiWatermarkTool binary before running it, and do not use the Feishu sharing snippet unless you intentionally want to share the image and have replaced the hard-coded recipient with the correct one.

Findings (3)

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.

Medium

#ASI03: Identity and Privilege Abuse

What this means

The agent can operate Gemini as the logged-in Google user, and prompts or generated images may be associated with that account.

Why it was flagged

The skill explicitly requires use of an already logged-in Google browser profile, allowing the agent to act through the user's authenticated session.

Skill content

- 浏览器已登录 Google 账号
- profile 使用 `user` 连接已打开的 Chrome

Recommendation

Use a dedicated browser profile or account if possible, keep the browser visible, and only allow the skill to operate on the intended Gemini pages.

Medium

#ASI04: Agentic Supply Chain Vulnerabilities

What this means

Installing or running the watermark tool gives third-party code local execution access on the user's machine.

Why it was flagged

The watermark-removal step depends on an external executable that is not included in the reviewed skill artifacts or pinned to a specific reviewed version.

Skill content

brew install allenk/tap/gwt ... 或从 [GitHub Releases](https://github.com/allenk/GeminiWatermarkTool/releases) 下载二进制文件 ... ~/.claude/skills/gwt/bin/GeminiWatermarkTool

Recommendation

Install the tool only from a trusted release, verify version/checksum where possible, and avoid running unknown preexisting binaries from unrelated skill directories.

Medium

#ASI07: Insecure Inter-Agent Communication

What this means

If used as written, the generated or processed image could be sent to an unintended Feishu recipient outside the Gemini workflow.

Why it was flagged

The optional sharing step imports another local skill and sends the image to a concrete Feishu user ID, rather than a placeholder or explicitly user-selected recipient.

Skill content

### 7. 发送到飞书（可选） ... send_image(
    image_path="/path/to/output.png",
    user_id="ou_7abe0c2af8a0f7b5b1c1171bcd8707d8"

Recommendation

Do not run the Feishu snippet unless the user explicitly requests sharing, verifies the recipient, replaces the hard-coded user ID, and trusts the separate send-feishu-image skill.