Skills
▥AGENT LAB: SKILLS
skills/op7418/document-illustrator-skill/document-illustrator/Gen Agent Trust Hub

document-illustrator

Pass

Audited by Gen Agent Trust Hub on Feb 17, 2026

Risk Level: SAFEPROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
  • [Indirect Prompt Injection] (LOW): The script scripts/generate_single_image.py interpolates user-provided text directly into a prompt for the Gemini API.\n
  • Ingestion points: CLI arguments content and title in scripts/generate_single_image.py.\n
  • Boundary markers: Absent. The script uses simple text headers which do not prevent subversion.\n
  • Capability inventory: Uses the google-genai library to communicate with an external API and writes image files to the local disk.\n
  • Sanitization: None. Input is used exactly as provided.\n- [Data Exfiltration] (LOW): The --style-file argument in scripts/generate_single_image.py allows reading local files and sending their content to an external service. If an attacker influences the path provided to this argument, sensitive local data could be exfiltrated to the API.\n- [Data Exfiltration] (SAFE): The script writes output to a user-defined path via the --output argument. While it can overwrite files, this is considered intended behavior for a generation tool.
Audit Metadata
Risk Level
SAFE
Analyzed
Feb 17, 2026, 04:45 PM