qwen-image-2

Pass

Audited by Gen Agent Trust Hub on Sep 2, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill instructions prompt the user to install the belt-sh/cli package and additional skills from the inference-sh organization on GitHub. These tools are required to interact with the underlying Alibaba models and the inference.sh platform.
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the user that is then passed to a high-capability inference tool, creating a potential vector for indirect prompt injection.
  • Ingestion points: Untrusted data enters the agent context through the prompt and reference_images (image URIs) parameters in the skill's example workflows.
  • Boundary markers: The instructions do not define explicit delimiters or "ignore embedded instructions" warnings for the prompt or reference_images content.
  • Capability inventory: The skill is granted access to the Bash tool (specifically the belt command) to execute image generation tasks and perform network operations via the CLI.
  • Sanitization: There is no evidence of input validation, URL filtering, or escaping for the user-provided prompt strings or reference image URIs before they are processed by the CLI.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 2, 2026, 02:18 PM
Security Audit — agent-trust-hub — qwen-image-2