huashu-design-html-native

Warn

Audited by Gen Agent Trust Hub on Sep 12, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONPRIVILEGE_ESCALATION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill instructs the user to install components using npx skills add alchaincyf/huashu-design. The source organization/user 'alchaincyf' is not verified and originates from an external registry.
  • [INDIRECT_PROMPT_INJECTION]: The 'Brand Asset Protocol' instructs the agent to search and download HTML and SVG assets from arbitrary external domains (e.g., <brand>.com/press) to extract brand colors.
  • Ingestion points: External brand websites and press kit pages defined by the user at runtime.
  • Boundary markers: None identified. The skill does not provide delimiters or instructions to ignore embedded content in the downloaded assets.
  • Capability inventory: The skill includes scripts for file writing (brand-spec.md), video rendering via Playwright, and shell script execution (scripts/convert-formats.sh).
  • Sanitization: None provided. The agent is instructed to run grep directly on the downloaded content, which could lead to following malicious instructions hidden in HTML/SVG metadata or comments.
  • [COMMAND_EXECUTION]: The skill utilizes a complex pipeline of shell scripts (convert-formats.sh, add-music.sh) and Node.js scripts (render-video.js) to generate MP4s and PPTX files, creating a significant execution surface for downloaded content.
  • [PRIVILEGE_ESCALATION]: The troubleshooting section suggests running apt install ffmpeg, which typically requires administrative (sudo) privileges on Linux systems.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Sep 12, 2026, 11:44 PM
Security Audit — agent-trust-hub — huashu-design-html-native