The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/detect-compute.py utility uses the subprocess module to probe system hardware and software capabilities. It executes standard commands such as nvidia-smi, nvcc, sysctl, and df to identify available VRAM and CPU resources. These operations are performed using list-based arguments without the shell, minimizing injection risks, and are necessary for the skill to self-constrain its machine learning recommendations based on available compute.
[DYNAMIC_EXECUTION]: The statistical utilities scripts/model-comparison.py and scripts/assumption-diagnostics.py utilize the statsmodels library to fit models based on R-style formulas provided as input. This enables the agent to perform flexible regression analyses on user-provided data. While this represents a dynamic evaluation surface, it is a standard implementation for statistical tooling in data science environments.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external data (CSV files) and user-provided statistical formulas.
Ingestion points: Data entering via scripts/assumption-diagnostics.py, scripts/model-comparison.py, and scripts/experimental-design.py using the --data flag.
Boundary markers: The skill relies on standard CSV parsing via pandas and formula parsing via patsy (internal to statsmodels).
Capability inventory: The skill possesses capabilities for subprocess execution (hardware probing) and file writing (exporting randomization schedules).
Sanitization: The skill follows standard library practices for data ingestion and provides comprehensive documentation in references/docker-experiment-isolation.md and references/subagent-experiment-supervision.md regarding safe execution and resource limits.

data-scientist