download-gemini-images
Pass
Audited by Gen Agent Trust Hub on Sep 7, 2026
Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses a bash script (
scripts/package_images.sh) to archive the collected images. This script executes the standardzipandzipinfoutilities to create and verify the integrity of the resulting archive.\n- [DYNAMIC_EXECUTION]: The skill dynamically loads a local JavaScript module (scripts/download_gemini_images.mjs) into the agent's Node.js environment. It also utilizes Playwright'sevaluatefunction to execute logic within the context of the Google Gemini web page for image extraction.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes content from an external source (the Gemini conversation page). Evidence chain:\n - Ingestion points: Browser tab DOM content from
gemini.google.com.\n - Boundary markers: None explicitly defined, but interaction is constrained by script logic.\n
- Capability inventory: File system write access, directory creation, and shell command execution (
zip).\n - Sanitization: The skill uses precise HTML element selectors to identify download buttons and images, which effectively isolates the automation logic from arbitrary text instructions contained within the conversation.
Audit Metadata