diffusers-cli
Pass
Audited by Gen Agent Trust Hub on Jul 30, 2026
Risk Level: SAFEREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSCREDENTIALS_UNSAFE
Full Analysis
- Remote Repository Code Execution: The
--trust-remote-codeparameter allows the skill to execute custom code from Hugging Face Hub repositories, which is required for certain model architectures but involves running external logic. - Remote Execution Environment: The skill can offload tasks to an external Hugging Face Sandbox via the
--remoteflag, which involves network data transfer and remote command invocation. - Dynamic Package Installation: Through the
--dependenciesflag, the skill can install arbitrary Python packages during remote execution to satisfy specific model requirements. - Credential Management: The tool retrieves the
HF_TOKENfrom the local environment and forwards it to the remote sandbox to facilitate access to restricted models. - External Content Loading: The CLI accepts URLs for media inputs (images, video, audio) in its arguments, fetching this data from the network before processing.
- Indirect Instruction Processing: The skill ingests untrusted input. Ingestion points:
pipeline-kwargsJSON and media URLs defined inrun.md. Boundary markers: No explicit instruction delimiters are mentioned in the usage examples. Capability inventory: Includes shell command execution, remote code loading, package installation, and file system operations. Sanitization: Relies on JSON schema validation but does not explicitly filter natural language content within prompts.
Audit Metadata