image-optimization

Warn

Audited by Gen Agent Trust Hub on Sep 9, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The diagnostic shell command provided in the skill is vulnerable to argument injection. By using xargs -I {} curl ... {}, the script allows URLs extracted from external websites to be passed as arguments to curl. A malicious actor could craft a URL containing curl flags (e.g., -o or --output) to perform unauthorized file writes on the system running the command.\n- [INDIRECT_PROMPT_INJECTION]: The skill defines a workflow for ingesting and processing data from untrusted external websites, creating a surface for injection attacks.\n
  • Ingestion points: External HTML content fetched via curl in the 'Diagnostic commands' section (SKILL.md).\n
  • Boundary markers: None. The instructions do not include delimiters or warnings to ignore instructions embedded in the crawled content.\n
  • Capability inventory: Shell command execution (curl, grep, xargs, awk) and vision-based analysis of external image URLs.\n
  • Sanitization: The regex used to extract URLs ([^"]+) is insufficient to prevent the inclusion of malicious CLI flags or shell-sensitive characters.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Sep 9, 2026, 12:02 AM
Security Audit — agent-trust-hub — image-optimization