ui-design

Warn

Audited by Gen Agent Trust Hub on Nov 17, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill explicitly directs the agent to use shell utilities like curl, wget, and Invoke-WebRequest to download remote assets (images, icons, and fonts) into the local project directory. This approach bypasses standard managed tool interfaces for network access and allows fetching from arbitrary external URLs.\n- [COMMAND_EXECUTION]: The agent is instructed to run shell commands for both asset management and code quality auditing. This includes using grep to search for forbidden patterns within generated files and using network tools for file transfers.\n- [INDIRECT_PROMPT_INJECTION]: The skill is configured to ingest and follow design constraints from external files such as DESIGN.md or brand specification documents. These ingestion points represent an attack surface where maliciously crafted files could influence agent behavior.\n
  • Ingestion points: Project-level files such as DESIGN.md, brand specifications, and design-token files (referenced in the 'Respect brand or design-system overrides' section).\n
  • Boundary markers: Absent; there are no instructions to the agent to treat these external files as untrusted or to delimit their content to prevent instructional override.\n
  • Capability inventory: The skill allows shell command execution (curl, grep), file writing, and complex code generation.\n
  • Sanitization: No validation or sanitization logic is provided to ensure external design inputs do not contain malicious instructions.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Nov 17, 2026, 04:34 PM