seo-images

Fail

Audited by Gen Agent Trust Hub on Sep 23, 2026

Risk Level: HIGHPRIVILEGE_ESCALATIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [PRIVILEGE_ESCALATION]: The skill instructions include commands using sudo apt install for installing system dependencies such as libimage-exiftool-perl and webp.\n- [COMMAND_EXECUTION]: The skill performs extensive subprocess operations using tools like exiftool, cwebp, convert, and ffmpeg. These operations interpolate user-supplied or scraped data (e.g., alt text, IPTC metadata values) into shell commands, representing a command injection risk.\n- [EXTERNAL_DOWNLOADS]: The skill guides the agent to recommend or perform package installations from official system repositories via apt. These are documented neutrally as standard dependency management from trusted sources.\n- [DYNAMIC_EXECUTION]: The skill executes local Python scripts (parse_html.py, iptc_ai_label.py) from the plugin's internal path using a custom script runner.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted web content from user-supplied URLs and local image metadata, creating a vulnerability surface.\n
  • Ingestion points: External HTML content and image file metadata (IPTC/XMP/EXIF).\n
  • Boundary markers: None identified. There are no instructions for the agent to delimit or ignore embedded instructions within the processed data.\n
  • Capability inventory: File writes, network resource ingestion, and shell command execution across multiple tools.\n
  • Sanitization: No specific sanitization or validation of the extracted external content is mentioned before it is processed or used in shell command generation.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Sep 23, 2026, 11:35 PM
Security Audit — agent-trust-hub — seo-images