sglang
Pass
Audited by Gen Agent Trust Hub on Oct 23, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: Fetches performance-optimized kernels from the FlashInfer package registry (flashinfer.ai), which is a well-known resource for accelerating LLM inference workloads.\n- [REMOTE_CODE_EXECUTION]: Includes instructions for cloning and installing the framework directly from its official GitHub repository (sgl-project/sglang).\n- [INDIRECT_PROMPT_INJECTION]: The skill demonstrates patterns for building agents and extraction pipelines that interpolate user-provided data into prompts, which is a standard pattern for LLM applications.\n
- Ingestion points: The text, article, and user_query parameters in the Python snippets.\n
- Boundary markers: Relies on structural formatting and JSON schemas to constrain output.\n
- Capability inventory: Supports structured generation, multi-modal processing, and tool usage.\n
- Sanitization: Examples focus on functional implementation without explicit string escaping.
Audit Metadata