gpt-image-2
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill facilitates the installation of the
@runcomfy/clipackage from the npm registry and fetches skill configuration from theagentspace-so/runcomfy-skillsrepository on GitHub. - [COMMAND_EXECUTION]: Invokes the
runcomfycommand-line tool to submit generation prompts to the RunComfy Model API and download the resulting image files to a local directory. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user inputs in the form of text prompts and external image URLs, presenting a surface for indirect prompt injection.
- Ingestion points: Input schemas for the
text-to-imageandeditendpoints defined inSKILL.md. - Boundary markers: The implementation uses JSON-formatted input for CLI arguments to ensure user-provided strings are not interpreted as shell commands.
- Capability inventory: CLI operations include making outbound HTTPS requests to the
model-api.runcomfy.netendpoint and writing files to the local file system. - Sanitization: Employs JSON serialization of user input to mitigate common shell command injection vulnerabilities.
Audit Metadata