matlab-recognize-text
Pass
Audited by Gen Agent Trust Hub on Aug 14, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: No malicious patterns detected. The skill follows best practices for MATLAB development and OCR workflows, including deterministic skew measurement and formal validation metrics.
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted image data through MATLAB's OCR functions. This attack surface is mitigated by the mandatory diagnostic workflow: 1) Ingestion occurs via imread in the diagnostic phase; 2) Boundary markers are established by the agent performing a visual read and plan before code execution; 3) Capabilities are restricted to standard MATLAB execution tools; 4) Sanitization is achieved through evaluateOCR metrics (CER/WER) and mandatory human-in-the-loop checkpoints for low-confidence results.
- [EXTERNAL_DOWNLOADS]: The skill refers to official MATLAB support packages for deep learning models and language data. These are legitimate resources hosted by the vendor (MathWorks).
Audit Metadata