expert-design

Fail

Audited by Gen Agent Trust Hub on Oct 3, 2026

Risk Level: HIGHCOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [COMMAND_EXECUTION]: The script tools/design-full/scripts/pick.sh is vulnerable to Python code injection. It takes the user-provided style description and inserts it directly into a Python command string using shell interpolation ('${QUERY}'). An attacker can use single quotes to break out of the string and execute arbitrary Python code on the host.
  • [DYNAMIC_EXECUTION]: The design-full pick command generates and executes Python code at runtime based on unsanitized user input. This implementation of python3 -c allows for the arbitrary code execution described above.
  • [INDIRECT_PROMPT_INJECTION]: The skill's workflow treats the user-supplied 'Brief' as the sole contract for its tasks, without sanitizing the input or establishing boundary markers. This creates an attack surface where instructions embedded in a brief could be used to manipulate the agent's behavior or tool usage. Ingestion points: brief.md; Boundary markers: None; Capability inventory: File system writes, tool calls (smart-search, awk-img-gen); Sanitization: None.
  • [EXTERNAL_DOWNLOADS]: The skill documentation suggests importing design system files from an untrusted third-party GitHub repository (VoltAgent/awesome-design-md) and domain (getdesign.md), which may lead to the ingestion of malicious configurations.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Oct 3, 2026, 11:19 AM