subagent-context-isolation

To ensure unbiased agent validation in a refine loop, enforce context-isolated sub-agent invocations where the validator agent only receives the artifact file for review, without any access to the creator agent's reasoning or history.

Implement file-based communication between agents by mandating that all interactions occur solely through defined artifact files such as PRDs, plans, and feedback status files, preventing direct context sharing between sub-agents.

Validator agents show bias during PRD refinement loops when they have access to the creator's thought process or history, which is resolved by enforcing strict input-only access to the artifact file for each sub-agent role.

Yes, context isolation applies to both plan refine loops and PRD reviews by enforcing independent sub-agent invocations, ensuring the validator agent only sees the specific artifact file for objective review.

Context isolation for sub-agents is the strict separation of concerns between creator and validator agents in iterative refinement loops, needed when you require objective artifact validation without the validator being influenced by the creator's reasoning.

Yes, strict input-only access for each sub-agent role is required in artifact validation to enforce independent invocations and prevent the validator from seeing the creator's context or reasoning history.