matlab-uihtml-app-builder

Pass

Audited by Gen Agent Trust Hub on Oct 15, 2026

Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill includes a calculator implementation in the file SKILL.md (Example 1) that uses the eval() function to process strings received from the HTML interface. While the code includes a whitelist validation check (allowedChars = '0123456789+-*/.() '), the use of eval on user-controlled input remains a sensitive pattern.
  • [DYNAMIC_EXECUTION]: The skill demonstrates the dynamic evaluation of strings as executable MATLAB code through the eval function. It explicitly warns against using this pattern without strict sanitization and provides a validateExpression function template to enforce character whitelisting.
  • [INDIRECT_PROMPT_INJECTION]: The architecture described in SKILL.md enables a direct interface for external data (HTML/JavaScript) to influence MATLAB backend execution, representing an indirect prompt injection surface.
  • Ingestion points: Untrusted data enters the MATLAB environment through the HTMLEventReceivedFcn callback in calculator.m, visualizer.m, and formProcessor.m when processing event.HTMLEventData.
  • Boundary markers: The instructions mandate wrapping all event handlers in try-catch blocks and setting HTMLSource = 'trusted' for local files.
  • Capability inventory: The backend scripts possess capabilities including arbitrary code evaluation (eval), file system interaction via user-prompted export (uiputfile, exportgraphics), and standard output logging (fprintf).
  • Sanitization: The skill provides specific patterns for sanitization, including whitelist character checking in the calculator example and explicit field validation (e.g., email and age checks) in the form processor example.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 15, 2026, 01:34 PM