mindspeed-llm-train-profiler
Pass
Audited by Gen Agent Trust Hub on Oct 10, 2026
Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes system commands including
npu-smi infoto verify hardware availability and sources environment configuration scripts likeset_env.sh. Its primary function involves executing a generated shell script to run the training profiler. - [DYNAMIC_EXECUTION]: The skill dynamically generates a shell script at runtime. It reads the user's input training script and appends profiling-specific arguments (e.g.,
--profile,--profile-ranks,--profile-level) to thetorchruncommand string before executing the new file. - [INDIRECT_PROMPT_INJECTION]: The skill has a data ingestion surface where it processes a user-provided
.shscript. Although it performs basic path verification, the skill ultimately executes the content of this external file, which constitutes an attack surface if the input script contains malicious commands.
Audit Metadata