flux-2-klein
Pass
Audited by Gen Agent Trust Hub on Jul 13, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to execute the
runcomfyCLI tool to interact with the RunComfy API. This involves running shell commands with user-defined parameters for model selection and input configuration. - [EXTERNAL_DOWNLOADS]: The CLI tool is configured to download generated assets directly from the vendor's official domains, specifically
runcomfy.netandruncomfy.com, into a local directory specified by the user. - [PROMPT_INJECTION]: The skill exposes a surface for indirect prompt injection.
- Ingestion points: User-supplied prompts are accepted as inputs and interpolated into a JSON object used as a command-line argument for the
runcomfyCLI. - Boundary markers: Prompts are encapsulated within JSON strings, which provides a basic but bypassable boundary.
- Capability inventory: The skill utilizes local command execution and network access to perform its primary function of image generation and result retrieval.
- Sanitization: While the documentation notes that the CLI tool itself does not shell-expand the prompt content, the agent's process of building the shell command line string presents a potential injection vector if special characters (e.g., quotes, backticks, or semicolons) are included in the prompt without appropriate escaping.
Audit Metadata