The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides instructions to execute local shell scripts (e.g., bash test/train_performance_8p_base_fp32.sh) and system utilities like ps, tail, and pkill. These commands are standard for managing machine learning training processes in an NPU environment and do not involve remote execution or privilege escalation.
[INDIRECT_PROMPT_INJECTION]: The skill monitors external log files for training progress (e.g., tail -20 <log_file>), which serves as an ingestion point for potentially untrusted data. While boundary markers and sanitization are absent, the risk is minimized by the specialized nature of the training logs and the skill's intended use case for monitoring known workloads.

model-training