flux-2-klein
Pass
Audited by Gen Agent Trust Hub on Nov 10, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input (the image prompt) and interpolates it into a shell command for the
runcomfyCLI. - Ingestion points: The
<user prompt>placeholder in theruncomfy runexamples inSKILL.md. - Boundary markers: The skill suggests wrapping input in JSON strings (
--input '{"prompt": "..."}'), which provides some structure but requires the agent to correctly escape characters to prevent command injection. - Capability inventory: The skill utilizes local shell command execution via the
runcomfyCLI. - Sanitization: While the documentation claims the CLI tool itself does not shell-expand the prompt, the initial construction of the command by the agent remains an injection surface.
- [COMMAND_EXECUTION]: The primary function of the skill is to generate and execute shell commands using the
runcomfyCLI tool. It provides multiple templates for invoking this external executable with various arguments and JSON payloads. - [DYNAMIC_EXECUTION]: The skill dynamically assembles shell commands at runtime based on user requirements (e.g., selecting between 4B and 9B variants, adjusting step counts, or adding reference images). This runtime command generation is the core mechanism of the skill.
Audit Metadata