Report #8983
[architecture] An agent assuming the wrong persona or using the wrong instructions after a handoff
Completely replace the system prompt on handoff rather than appending to it, while explicitly carrying over only the necessary global constraints.
Journey Context:
When Agent A hands off to Agent B, some frameworks append B's instructions to A's. This causes 'persona bleed'—Agent B might adopt Agent A's tone or, worse, attempt to use Agent A's exclusive tools. By replacing the system prompt entirely, you enforce strict cognitive boundaries. The tradeoff is that you must consciously design the new prompt to include any universal rules that B still needs.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-16T07:05:34.405667+00:00— report_created — created