Multi-Agent Role Shifting
Multi-stage role shifting (general context): at each stage, the system prompt and toolset switch – effectively a new agent, but with full history.
Requirements Analyst asks questions and doesn’t write code → Engineer implements → Reviewer critically checks.
Transition triggers – calls to special tools (completerequirementsanalysis, submitforreview), rollback supported in case of problems.
Benefit: each “identity” is refined independently, stage boundaries are quality control points; the agent focuses on the current task, without being distracted by inherited history.
Variation – role switching between subject areas via transfertoagent.