matlab-ocr

Pass

Audited by Gen Agent Trust Hub on Aug 2, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill implements strong process guardrails, such as 'Critical Rule #1', which prevents users from overriding the mandatory diagnostic and planning phase, protecting against certain types of instruction bypass.
  • [SAFE]: The skill enforces a capability gate by requiring the agent to verify the existence of MATLAB support packages before execution and providing clear instructions to the user for manual installation, avoiding silent failures or unauthorized downloads.
  • [SAFE]: The workflow for processing untrusted image data includes a mandatory evidence chain for safety: ingestion occurs via imread in SKILL.md and referenced files; boundaries are established by 'Critical Rule #7' (writing files before reporting) and Step 5 (Confidence checkpoint); capabilities include MATLAB code evaluation and file writing via the toolset defined in manifest.yaml; and sanitization is managed through confidence filtering (WordConfidences > 0.7) and mandatory ground-truth validation via evaluateOCR.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 2, 2026, 12:11 PM
Security Audit — agent-trust-hub — matlab-ocr