vulkan-optimize
Pass
Audited by Gen Agent Trust Hub on Nov 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONCOMMAND_EXECUTION
Full Analysis
- [SAFE]: The skill contains instructional content for developers to optimize the MNN Vulkan backend. All operations, including local compilation, file manipulation, and device-side profiling, are consistent with standard software engineering and performance optimization practices for mobile platforms.
- [INDIRECT_PROMPT_INJECTION]: The skill identifies typical ingestion points for an AI agent, such as reading GLSL source files and performance logs. While this presents an attack surface for indirect prompt injection, it is managed through documentation of intended inputs and standard developer workflows. No exploitation patterns were observed.
- [DYNAMIC_EXECUTION]: The workflow describes the generation of SPIR-V binary code from GLSL shaders, which are then compiled into the final executable. This is an expected build-time step in GPU-accelerated software development.
- [COMMAND_EXECUTION]: The documentation references the use of standard command-line utilities such as
cmake,make, andadbfor building and testing libraries on target devices. These commands are used as intended for technical development and do not demonstrate suspicious behavior or unauthorized system access.
Audit Metadata