gemini-web-media-analysis
Pass
Audited by Gen Agent Trust Hub on Aug 15, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses a JavaScript tool to inject a helper file input element and simulate a clipboard paste event within the browser context, which is necessary to bypass standard interface limitations.
- [EXTERNAL_DOWNLOADS]: The skill facilitates uploading local media files to the
gemini.google.comdomain. As Google is a well-known service, this is considered intended functionality, and the skill includes an advisory to obtain user consent before uploading data. - [PROMPT_INJECTION]: The skill reads external data from the Gemini web interface, creating a vulnerability surface for indirect prompt injection.
- Ingestion points: The agent context is updated with untrusted data when polling the analysis result using the
get_page_texttool in SKILL.md. - Boundary markers: The instructions do not define any delimiters or warnings to ignore instructions contained within the retrieved page text.
- Capability inventory: The skill possesses extensive capabilities including browser navigation, JavaScript execution, and local file upload via the
claude-in-chromeserver. - Sanitization: No sanitization, filtering, or validation is performed on the data ingested from the browser session.
Audit Metadata