The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts scripts/verify_kernel.py and scripts/benchmark_kernel.py use subprocess.run to execute Python-based verification and benchmarking harnesses. This is a functional requirement for the skill, allowing it to programmatically test the GPU kernels developed by the agent.\n- [REMOTE_CODE_EXECUTION]: The skill uses Python's importlib.util in its testing scripts to dynamically load and execute the generated kernel code. This behavior is used exclusively for local validation of the agent's own output and does not involve downloading or executing code from untrusted remote sources.\n- [SAFE]: No indicators of credential theft, data exfiltration, or malicious prompt injection were identified. All external documentation links point to the official Triton language website, and the repository follows safe development practices.

kernel-triton-writing