The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from external CSV and Excel files, creating a potential surface for indirect prompt injection if the files contain instructions targeting the agent.
Ingestion points: Multiple functions across SKILL.md read data from files into the agent's memory using pandas, csv, and openpyxl libraries.
Boundary markers: The skill does not include specific prompt delimiters or instructions to the agent to disregard natural language instructions found within the data rows.
Capability inventory: The skill's environment includes tools for Bash and file system modification (Write, Edit), and the code snippets provide capabilities for database interaction via sqlalchemy.
Sanitization: The wrangle_csv pipeline includes a normalization step that strips non-alphanumeric characters from column names using regex (str.replace(r'[^\w]', '', regex=True)), which provides basic protection against schema-based confusion, though row data remains unsanitized.

csv-wrangler