huggingface-gradio
Pass
Audited by Gen Agent Trust Hub on Oct 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-provided Gradio code and fetches external information via SandBase APIs.
- Ingestion points: Gradio source code and external outputs from SandBase tools (SKILL.md, references/sandbase-api-map.md).
- Boundary markers: A 'SandBase boundary' is established to limit external data flow (SKILL.md).
- Capability inventory: The skill uses tools like sandbase_discover, sandbase_inspect, and sandbase_run for external interactions.
- Sanitization: Focus checks explicitly require validating uploaded files (SKILL.md).
Audit Metadata