The Agent Skills Directory

[SAFE]: The skill is a specialized debugging utility that automates the addition of telemetry to local machine learning model files. This instrumentation is used to compare numerical outputs between hardware devices and is consistent with the stated purpose of the tool.
[SAFE]: The use of the Python compile() function in inject_tensor_dump.py and rollback_tensor_dump.py is a safety measure used to verify that the programmatically modified model code is syntactically correct before it is saved or used.
[SAFE]: The tool captures only non-sensitive tensor metadata (e.g., L1 norms, shapes, and data types) and records this information to local logs using standard logging frameworks. No network exfiltration or unauthorized access to sensitive files was detected.
[SAFE]: The provided scripts for log analysis and code verification operate locally on the user's machine and do not download or execute code from untrusted external sources.

inference-precision-tensor-dump-compare