The Agent Skills Directory

[COMMAND_EXECUTION]: The script op_perf_analysis_combine.py utilizes subprocess.run to orchestrate multiple analysis steps by calling other scripts within the same package. This is a standard and secure pattern for tool chaining, as it passes arguments as a list and uses the current Python environment via sys.executable, which prevents shell command injection.
[EXTERNAL_DOWNLOADS]: The extract_op_shapes.py script includes logic to detect the presence of the openpyxl dependency and provides standard installation instructions to the user if it is missing. No automated or background downloads of external code were found.
[DATA_EXFILTRATION]: The skill's operations are strictly limited to the local filesystem, specifically reading profiling data from user-defined paths and writing analysis results to an output directory. No network-related functions or unauthorized access to sensitive system directories were detected.
[PROMPT_INJECTION]: The SKILL.md and associated documentation are focused exclusively on technical performance analysis instructions. There are no attempts to override model safety filters, extract system prompts, or implement jailbreak patterns.

profiling-analysis-computing