flux-2-klein
Pass
Audited by Gen Agent Trust Hub on Sep 23, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill instructions document the installation of the
@runcomfy/clipackage from the NPM registry and the skill repository from GitHub (agentspace-so/runcomfy-skills). These are necessary components for the model interface described. - [COMMAND_EXECUTION]: The skill operates by executing the
runcomfyCLI tool to interact with remote inference endpoints. It provides specific command templates for various model variants (4B and 9B). - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input (prompts) which are passed through a local CLI to a remote AI model, creating a potential vector for instructions embedded within the data to influence model behavior.
- Ingestion points: User-supplied strings passed to the
promptfield in the CLI commands documented in SKILL.md. - Boundary markers: Input is encapsulated as a value within a JSON object literal passed to the
--inputflag. - Capability inventory: The skill utilizes shell command execution and network communication via the external RunComfy CLI.
- Sanitization: Documentation indicates the CLI transmits input as a JSON body over HTTPS without shell expansion to mitigate command injection at the shell level.
Audit Metadata