simulink-streaming-dsp
Pass
Audited by Gen Agent Trust Hub on Aug 30, 2026
Risk Level: SAFE
Full Analysis
- [PROMPT_INJECTION]: No evidence of prompt injection or instructions to bypass safety filters was found. The skill uses neutral, technical language focused on Simulink model construction.
- [DATA_EXFILTRATION]: No patterns associated with data exfiltration or access to sensitive local files (e.g., credentials, SSH keys) were detected.
- [OBFUSCATION]: No obfuscated code, Base64 strings, zero-width characters, or homoglyph attacks were identified.
- [REMOTE_CODE_EXECUTION]: The skill does not perform external downloads or execute code from untrusted remote sources. It relies on the local MATLAB environment and specific MCP tools for model manipulation.
- [COMMAND_EXECUTION]: While the skill uses
evaluate_matlab_code, it provides strict instructions to limit its use to model verification and mathematical calculations, explicitly prohibiting its use for model construction in favor of safer MCP tools likemodel_edit. - [DYNAMIC_CONTEXT_INJECTION]: No use of dynamic context injection (the !command pattern) was found in the SKILL.md file.
- [INDIRECT_PROMPT_INJECTION]: The skill processes Simulink model metadata and signal configurations, but does not ingest untrusted external content in a way that creates an injection surface.
- [METADATA_POISONING]: Metadata fields correctly reflect the author (MathWorks) and the skill's technical purpose.
Audit Metadata