The Agent Skills Directory

[PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data (CSV files, SQL databases), which constitutes a surface for indirect prompt injection attacks.
Ingestion points: Data ingestion occurs through pd.read_csv('data.csv') and SQL SELECT queries as described in the loading and exploration step.
Boundary markers: The instructions lack explicit delimiters or warnings instructing the agent to ignore any natural language instructions that might be embedded within the datasets.
Capability inventory: The skill has access to the Bash tool and file system via Read, Grep, and Glob, which increases the potential impact if an injection attack successfully influences agent actions.
Sanitization: No data validation or sanitization of the loaded datasets is performed in the provided code templates.

data-analysis