The Agent Skills Directory

[SAFE]: The skill establishes a structured environment for financial modeling and reinforcement learning diagnostics. All components, including the Python scripts and markdown references, follow standard data science practices without introducing security risks.
[COMMAND_EXECUTION]: The skill executes a local Python script reinforcement_learning_trading_diagnostics.py. Analysis of the script confirms it performs deterministic statistical calculations (mean, standard deviation, quantiles) using the pandas library and does not invoke shell commands or interact with the network.
[INDIRECT_PROMPT_INJECTION]: The workflow involves ingesting external data via CSV files.
Ingestion points: The scripts/reinforcement_learning_trading_diagnostics.py script reads input.csv as specified in SKILL.md.
Boundary markers: The script enforces a specific schema of 12 required columns, which acts as a structural validation layer.
Capability inventory: The skill is limited to reading CSV files and writing JSON diagnostic artifacts; it lacks network access, arbitrary code execution, or elevated filesystem privileges.
Sanitization: The script uses pd.to_numeric and pd.to_datetime with errors="coerce", which effectively sanitizes input by converting non-conforming data to NaN/NaT rather than executing it.

reinforcement-learning-trading