objective-research

To prevent confirmation bias during codebase research, separate agent contexts for intent and investigation. This ensures research questions remain factual and intent-free, yielding objective data on existing code paths before development.

To analyze existing data flow and call paths, use a two-context agentic workflow to decompose tickets into targeted research questions. This gathers objective data on current behavior and feature implementations before you start coding.

A factual investigation workflow for unfamiliar code paths splits intent from investigation using two distinct agent contexts. This mechanism ensures the research agent gathers objective data without knowing the ultimate development goal.

To decompose implementation tickets into specific research questions, apply an intent-free context to the ticket details. This generates targeted, factual queries about existing code paths and current behavior for objective codebase investigation.

Use a two-context workflow for code research when you need to implement a new feature in an unfamiliar part of the codebase. It prevents the research agent's confirmation bias by keeping the ultimate goal hidden during data gathering.