The Agent Skills Directory

[COMMAND_EXECUTION]: The skill contains a system probe script (scripts/detect-compute.py) and a supervision reference (references/subagent-experiment-supervision.md) that use subprocess to execute commands.
scripts/detect-compute.py executes nvidia-smi, nvcc, sysctl, and df to inventory available hardware resources and ML environment capabilities.
The supervision documentation describes a pattern for an automated supervisor that can execute experiment scripts and apply fixes by modifying command-line arguments (e.g., reducing batch size) or installing missing Python modules via pip install when an ImportError is detected.
[DATA_EXFILTRATION]: The scripts/detect-compute.py script collects system metadata including CPU/GPU specifications, total/available RAM, and disk space. This information is intended to help the agent self-constrain its modeling approach and is output to the local environment only; no network exfiltration patterns were detected. The references/subagent-experiment-supervision.md file contains a Telegram notification template for experiment alerts, but it uses placeholders and is not configured for automatic execution.
[EXTERNAL_DOWNLOADS]: The skill's documentation and Docker configuration reference official scientific libraries and well-known repositories, including PyTorch, Scikit-Learn, and various GitHub organizations. These references follow standard development practices and do not involve unverified remote script execution.
[SAFE]: The skill's behavior aligns with its stated purpose as a PhD-level data science assistant. The use of system probes and process management is a functional requirement for managing complex machine learning experiments in a development environment.

data-scientist