Security audit
Security checks for vulnerabilities and agentic risk
This skill is a coherent Gemini web automation helper, but users should understand that their prompts and generated images go through Gemini and may use a logged-in browser profile.
Install only if you are comfortable having selected prompts sent to Gemini through the OpenClaw browser profile. Avoid sending secrets or sensitive personal data, and clean up downloaded generated images if they should not remain on disk.
The image-generation trigger list includes the standalone word "画", which is a very common everyday term in Chinese and can appear in requests unrelated to image generation. This can cause the skill to activate unexpectedly and send user content to Gemini without clear intent, creating privacy and unintended external-action risks.
The workflow instructs the agent to send user prompts directly to gemini.google.com and depend on an authenticated session, but it does not require a user-facing disclosure or confirmation before transmitting potentially sensitive content. This is dangerous because users may not realize their data is being shared with a third-party service under the operator's logged-in account.
The flow explicitly instructs the agent to download generated images to local storage and then send them back through another channel, but it provides no guidance on user consent, retention limits, deletion, or handling of potentially sensitive generated content. This creates a real data-handling risk because local downloads can persist on disk, leak across sessions, or expose private/generated material beyond the minimum needed to complete the task.
The natural-language instructions and suggested user messages are all presented in Chinese, which may effectively force a specific language or locale for the interaction. There is no stated option for the user to choose another language or confirmation that the skill is intentionally limited to Chinese-speaking users.
The selector set explicitly targets Chinese strings like "发送" and "图片" alongside English strings, which embeds language assumptions into the skill behavior. There is no natural-language indication that the skill is limited to specific locales or that users can choose supported languages, which can violate locale-choice policy expectations.
No suspicious patterns detected.
