charles-color-analysis

Pass

Audited by Gen Agent Trust Hub on Nov 16, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [COMMAND_EXECUTION]: The Python script scripts/color.py executes external tools and scripts using the subprocess module. Specifically, it invokes a Swift script (scripts/ocr.swift) to perform Optical Character Recognition (OCR) on images using macOS Vision APIs and controls Chrome or Chromium in headless mode to render HTML files.
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from user-provided images and HTML documents. Text extracted from these sources via OCR is included in the analysis output and presented to the AI agent. There are no explicit boundary markers or sanitization steps to prevent malicious instructions embedded within these images or pages from influencing the agent's behavior.
  • Ingestion points: scripts/color.py reads image files (processed by ocr.swift) and HTML files.
  • Boundary markers: Absent in the instructions for handling the extracted text content.
  • Capability inventory: The skill can execute local commands, write to the filesystem, and interact with a browser engine.
  • Sanitization: No sanitization is applied to the text recognized by the OCR tool or the content within HTML files.
  • [DYNAMIC_EXECUTION]: The shot command in scripts/color.py performs dynamic script injection. It modifies a local HTML file by appending a JavaScript file (scripts/dom_colors.js) to the document body before rendering it in a headless browser to extract color data and capture screenshots.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 16, 2026, 05:26 PM