The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill interacts with external web applications and captures visual evidence for AI review, creating a potential surface for indirect prompt injection where malicious content on a target site might influence the Gemini model's output. Ingestion points include browser content and screenshots accessed via platform tools; the review script in resources/gemini-review.md includes instructional framing but lacks explicit separators for untrusted visual content; capabilities include browser automation, file writing, and network requests to the Gemini API; and no specific filtering is applied to the captured visual data before processing.- [DYNAMIC_EXECUTION]: The skill provides a Node.js script template in resources/gemini-review.md that performs local execution to interface with the Gemini API. While serving a legitimate functional purpose, it involves the execution of provided script content.- [EXTERNAL_DOWNLOADS]: The skill requires the @google/generative-ai Node.js package, which is an official and trusted library from a well-known service provider.

visual-qa