matlab-use-experiment-manager
Pass
Audited by Gen Agent Trust Hub on Oct 1, 2026
Risk Level: SAFECOMMAND_EXECUTIONOBFUSCATIONPROMPT_INJECTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses the
evaluate_matlab_codetool to perform MATLAB operations, including writing script files to the project directory and invoking helper functions for experiment management. - [OBFUSCATION]: Several core logic components are provided as MATLAB P-files (e.g.,
AppController.p). These are pre-parsed, non-human-readable files used by the vendor to distribute code securely while protecting intellectual property. - [PROMPT_INJECTION]: The instruction set in
common.mddirects the agent to suppress the narration of internal API calls and helper function names. While this limits technical transparency for the user, it is a standard design pattern for maintaining a streamlined agent persona. - [INDIRECT_PROMPT_INJECTION]: Because the skill generates MATLAB code based on user-provided descriptions, it possesses an attack surface for indirect prompt injection. The skill mitigates this through mandatory multi-tier validation checks, including static analysis with
checkcodeand function signature verification.
Audit Metadata