localmac-ai-ocr

Pass

Audited by Gen Agent Trust Hub on Oct 8, 2026

Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses subprocess.run within scripts/gui_toolkit.py to execute macOS system utilities including osascript, screencapture, and sips. These are used for window management, taking screenshots, and image processing respectively.
  • [DYNAMIC_EXECUTION]: The script scripts/gui_toolkit.py uses ctypes.CDLL to dynamically load the macOS system framework /System/Library/Frameworks/ApplicationServices.framework/ApplicationServices. This is utilized to perform low-level mouse events (clicks) through the Core Graphics API. Additionally, the skill constructs and executes AppleScript strings at runtime to interact with the macOS user interface.
  • [INDIRECT_PROMPT_INJECTION]: The skill represents a surface for indirect prompt injection through its OCR workflow.
  • Ingestion points: Text is extracted from screenshots via the scripts/ocr_tool.py script and returned to the agent.
  • Boundary markers: The skill output does not explicitly wrap extracted text in protective delimiters before passing it back to the agent context.
  • Capability inventory: The skill possesses capabilities to perform file system writes (screenshots in /tmp), network requests (to the configured OCR API), and GUI actions like clicking and typing via gui_toolkit.py.
  • Sanitization: There is no evidence of sanitizing or filtering the text recognized by the OCR engine before it is processed by the agent.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 8, 2026, 11:42 PM