image-editor

Pass

Audited by Gen Agent Trust Hub on Aug 6, 2026

Risk Level: SAFEDATA_EXFILTRATIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [DATA_EXFILTRATION]: The skill uses a background Python script to monitor the system clipboard using the pbpaste utility. While intended to transfer data from the React editor sandbox to the processing script, this provides the skill with access to all user clipboard content during the 5-minute polling window. The data is stored locally in /tmp/editor_apply_ready.json.
  • [COMMAND_EXECUTION]: The skill executes the pbpaste system command via Python's subprocess.run to retrieve clipboard data and uses a tool to display the final image.
  • [PROMPT_INJECTION]: The skill ingests untrusted data from the system clipboard. Although it validates the JSON schema, the content is used directly in image rendering tasks.
  • Ingestion points: macOS clipboard via pbpaste (SKILL.md).
  • Boundary markers: Schema validation for specific JSON keys (headline, brand, logo, gradient).
  • Capability inventory: Clipboard access, local file writing to /tmp, image compositing with Pillow, and file opening.
  • Sanitization: Basic JSON structure validation is performed; however, text content from the clipboard is rendered onto images without additional sanitization.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 6, 2026, 06:00 AM
Security Audit — agent-trust-hub — image-editor