comfyui-api
Fail
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: HIGHEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill uses
curlto download generated output files from the ComfyUI server. By default, these operations target the local loopback address (127.0.0.1:8188), which is the standard operational procedure for a local ComfyUI setup. - [COMMAND_EXECUTION]: The documentation provides various
curlcommands to interact with the service, such as checking system status, queuing prompts, and uploading images. These commands are well-scoped to the intended administrative and creative functions of the ComfyUI API. - [INDIRECT_PROMPT_INJECTION]: The skill reads data from local files (
state/inventory.json,references/models.md) to perform workflow validation. This creates an indirect injection surface if these files are compromised. - Ingestion points: Workflow node validation and model availability checks read from local JSON and Markdown files.
- Boundary markers: No specific delimiters or instructions to ignore embedded content are used during the file processing.
- Capability inventory: The skill can execute network requests via
curland write to the local file system (saving workflows and downloading images). - Sanitization: There is no evidence of strict schema validation or sanitization of the content read from the inventory or reference files.
Recommendations
- HIGH: Downloads and executes remote code from: http://127.0.0.1:8188/view?filename=ComfyUI_00001.png&subfolder=&type=output - DO NOT USE without thorough review
Audit Metadata