The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides code templates that are vulnerable to SQL injection. Specifically, the incremental_by_timestamp and full_refresh_with_swap functions interpolate variable table and column names directly into SQL strings using Python f-strings. This pattern is dangerous if the identifiers are sourced from untrusted external data (e.g., user input or file names).
Ingestion points: The functions accept table and timestamp_col as parameters in SKILL.md which an agent might populate with external data.
Boundary markers: No delimiters or warnings are provided to prevent the agent from treating untrusted data as SQL identifiers.
Capability inventory: The templates utilize pandas.read_sql and sqlalchemy to execute the resulting queries against a database.
Sanitization: The patterns lack identifier quoting or validation, which is necessary when parameterizing SQL identifiers.

etl-incremental-patterns