The Agent Skills Directory

[COMMAND_EXECUTION]: The skill guides the agent to execute several shell commands essential for its purpose, including building operators with cmake, make, and bash, and running the msprof NPU profiling tool. These are standard developer tools in the CANN (Compute Architecture for Neural Networks) environment.
[DATA_EXFILTRATION]: The skill involves reading profiling data (CSV files) and writing statistical summaries and archives to the local filesystem. These operations are performed by the included scripts/perf_summary.py script and are restricted to the project's operator directory. No network operations or external data transmissions were identified.
[PROMPT_INJECTION]: The skill processes data from profiling outputs which could be a vector for indirect prompt injection. However, the processing is limited to statistical aggregation (min, max, average) performed by a local Python script, and the resulting summary is used by the agent for performance evaluation. This is a standard data analysis pattern and does not represent a significant risk.

external-cannbot-ops-ops-profiling