matlab-use-package-manager

Pass

Audited by Gen Agent Trust Hub on Oct 10, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill has an attack surface where untrusted metadata—such as package names, summaries, and versions—from external search endpoints and custom repositories is ingested and used as arguments in package management commands.
  • Ingestion points: Metadata is retrieved from the MathWorks File Exchange search endpoint (www.mathworks.com/searchresults/results) and custom Repository Server REST APIs (v1/packages/index.json).
  • Boundary markers: Absent; the skill does not implement delimiters or specific warnings to the agent to ignore instructions embedded in the retrieved metadata.
  • Capability inventory: The skill uses mpminstall and mpmupdate to install and run code, facilitated by the evaluate_matlab_code tool.
  • Sanitization: No programmatic sanitization is applied to ingested metadata; the skill relies on manual mitigations such as DryRun and user confirmation.
  • [EXTERNAL_DOWNLOADS]: The skill performs network requests to discover and download package metadata from external sources.
  • Evidence: Uses curl to fetch data from www.mathworks.com and user-configured repository URLs, and webread for internal MATLAB network operations.
  • [COMMAND_EXECUTION]: The skill executes MATLAB commands that modify the system state, including installing, updating, and removing software packages.
  • Evidence: Implements the mpm command family (mpminstall, mpmupdate, mpmuninstall) which impacts the file system and MATLAB path.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 10, 2026, 09:13 AM