The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion and processing of external data sources, creating a potential surface for indirect prompt injection if the processed files are attacker-controlled.
Ingestion points: External data is loaded using pd.read_csv, pd.read_parquet, and pd.read_feather across multiple reference files, including references/data-cleaning.md and references/performance-optimization.md.
Boundary markers: The skill instructs the agent to validate data quality using assert checks in SKILL.md and provides patterns for schema validation using the pandera library in references/data-cleaning.md.
Capability inventory: The agent is granted file system read and write capabilities through the pandas library, including the use of pd.eval() for optimized expression evaluation as shown in references/performance-optimization.md.
Sanitization: The skill provides extensive documentation for sanitizing data, including missing value handling (fillna, dropna), strict type enforcement (astype), and regex-based string validation in references/data-cleaning.md.

pandas-pro