Skills
Navigate
skills/jiangmuran/claude-image/gpt-image-2/Gen Agent Trust Hub

gpt-image-2

Warn

Audited by Gen Agent Trust Hub on May 4, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONCREDENTIALS_UNSAFEEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The install.sh script modifies the user's shell configuration (~/.zshrc or ~/.bashrc) by appending export statements, which persists environment variables across sessions.
  • [CREDENTIALS_UNSAFE]: The installation process stores the OPENAI_IMAGE_API_KEY in plaintext within the user's shell profile files, increasing the risk of credential exposure.
  • [EXTERNAL_DOWNLOADS]: The scripts/gpt_image.py script communicates with and downloads data from an external API endpoint (https://jmrai.net/v1) using the urllib library.
  • [PROMPT_INJECTION]: The skill implements a 'visual self-verification' loop in SKILL.md where the agent is instructed to read generated PNG files (ingestion point: scripts/gpt_image.py output) and iterate based on their content. This workflow lacks boundary markers or sanitization, creating an indirect prompt injection surface where adversarial content in images could influence the agent's future shell commands (capability: python3 subprocess execution).
Audit Metadata
Risk Level
MEDIUM
Analyzed
May 4, 2026, 02:30 PM
Security Audit — agent-trust-hub — gpt-image-2