gpt-image-skill
Pass
Audited by Gen Agent Trust Hub on Sep 30, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
- [COMMAND_EXECUTION]: The skill relies on the execution of a bundled Python script
gpt_image.pyto perform its primary functions. The script handles argument parsing, API interaction, and file I/O. - [DATA_EXPOSURE]: The script
gpt_image.pyis configured to read sensitive credentials (OPENAI_API_KEY) from~/.gpt-image.env. While this is a common practice for CLI tools, it involves accessing a sensitive path in the user's home directory. - [INDIRECT_PROMPT_INJECTION]: The instructions in
SKILL.mddirect the agent to interpolate user-provided prompts directly into shell command strings. This creates a potential surface for command injection if the underlying agent platform does not sanitize or escape shell metacharacters before execution. - Ingestion points: User-provided image prompts and filenames in
SKILL.md. - Boundary markers: Prompts are enclosed in double quotes within the shell command examples in
SKILL.md. - Capability inventory: The skill can execute Python scripts, write images to the local filesystem, and make outbound network requests to OpenAI and other URLs via the
httpxlibrary. - Sanitization: No explicit sanitization logic for user input is present in the markdown instructions.
Audit Metadata