visual-style
Pass
Audited by Gen Agent Trust Hub on Nov 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by ingesting and processing untrusted data from external websites, PDFs, and videos.\n
- Ingestion points: External untrusted content is ingested via website URLs (HTML/CSS), uploaded PDF brand guides, and video files/URLs as described in
references/extractors/from-website.md,references/extractors/from-pdf.md, andreferences/extractors/from-video.md.\n - Boundary markers: No specific delimiters or "ignore embedded instructions" warnings are used when interpolating extracted content into the
visual-style.mdYAML fields or the subsequent prompts generated for downstream tools.\n - Capability inventory: The skill generates natural language prompts for the HeyGen Video Agent (
references/connectors/heygen-video-agent.md) and produces HTML/CSS code (references/connectors/html-slides.md), both of which can be influenced by malicious instructions embedded in the source data.\n - Sanitization: There is no evidence of sanitization, escaping, or validation of the text extracted from external sources before it is used to populate critical fields like
style_prompt_full.
Audit Metadata