image-gen

Fail

Audited by Gen Agent Trust Hub on Mar 10, 2026

Risk Level: HIGHCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill is vulnerable to shell command injection in SKILL.md during the Ideogram generation process. User-controlled variables, including the image prompt ($PROMPT) and reference image paths ($CHAR_REF_PATH, $STYLE_REF_PATH), are interpolated into a command string that is executed via the eval command. This allows an attacker to execute arbitrary shell commands by including metacharacters like semicolons, backticks, or pipes in their input.\n- [EXTERNAL_DOWNLOADS]: The skill interacts with external APIs at api.ideogram.ai and generativelanguage.googleapis.com to generate images. It programmatically downloads and saves these files to the local filesystem at ~/image-gen/ and opens them automatically using the system's open command.\n- [PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface because it ingests untrusted user data and uses it in operations with significant system capabilities without proper sanitization or boundary markers.\n
  • Ingestion points: User-provided prompts and file paths from command arguments in SKILL.md.\n
  • Boundary markers: Absent; inputs are directly interpolated into shell commands and JSON request bodies.\n
  • Capability inventory: Execution of shell commands (curl, mkdir, eval), file writing (python3), and opening local files.\n
  • Sanitization: No escaping or validation is performed on user inputs before they are used in high-privilege operations.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Mar 10, 2026, 03:36 AM