concept-to-image

Pass

Audited by Gen Agent Trust Hub on Mar 7, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection where untrusted user input (the 'concept') is used to generate HTML content that is subsequently rendered in a browser.\n
  • Ingestion points: User concept input processed in SKILL.md (Step 1) and transformed into HTML.\n
  • Boundary markers: SKILL.md includes instructions to keep HTML self-contained with no external resources, which serves as a behavioral constraint but not a technical boundary.\n
  • Capability inventory: scripts/render_to_image.py executes a headless Chromium process using Playwright to render and screenshot HTML artifacts.\n
  • Sanitization: Absent; the Python rendering script does not sanitize or validate the HTML content before execution.\n- [COMMAND_EXECUTION]: The skill invokes a local Python script scripts/render_to_image.py to handle the rendering process. This script launches a browser process with the --no-sandbox flag, which disables significant security isolation features of the Chromium browser.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 7, 2026, 03:56 PM