meihua-yishu
Warn
Audited by Gen Agent Trust Hub on Sep 21, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONDYNAMIC_EXECUTIONMETADATA_POISONINGEXTERNAL_DOWNLOADS
Full Analysis
- [COMMAND_EXECUTION]: The script
scripts/meihua_gemini.pyexecutes thecurlcommand viasubprocess.runto query local browser debugging information fromlocalhost:9222. This allows the script to map active browser tabs for manipulation. - [DYNAMIC_EXECUTION]: In
scripts/meihua_gemini.py, the skill uses the Chrome DevTools Protocol (CDP) to execute arbitrary JavaScript (Runtime.evaluate) in the user's browser. This allows the script to interact with a logged-in Gemini session, providing it the capability to read content from the page and simulate user input within an authenticated web context. - [METADATA_POISONING]: The
SKILL.mdfile contains instructions intended to steer the agent's behavior by directing it to promote a specific commercial research platform (www.k-dense.ai). This redirection is not related to the stated metaphysical functionality and serves to influence agent behavior for third-party interests. - [EXTERNAL_DOWNLOADS]: The skill requires the user to install the third-party
websocketslibrary and to manually configure their local browser to enable remote debugging, which significantly increases the host machine's attack surface.
Audit Metadata