A Data Extraction parent agent must map an unfamiliar 600-table warehouse before generating queries. Probing each table's schema floods the parent's context with verbose DDL it will never reuse, yet auditors later require proof of which columns informed each query. Which design keeps the parent lean while preserving an auditable record?
A. Have the parent probe every table itself and retain all raw DDL in context.
B. Delegate probing to a subagent but have it return the complete DDL for every probed table to the parent.
C. Delegate probing to a subagent that returns only relevant columns and writes the inspected DDL to a scratchpad file.
D. Skip the subagent and let the parent compact its history when the DDL no longer fits.
Drop your answer (A / B / C / D) in the comments 👇 I'll reveal the correct answer and the why tomorrow.