My review agent audited the wrong subsystem for $1.35 because of a 3-month-old session id

A developer encountered an issue where their review agent audited the wrong subsystem due to a 3-month-old session ID, costing $1.35. The system dispatches a reviewer and security auditor on a different provider, using models like codex/gpt-5.6. Although the review cost $0.92 and generated 50KB of text, the findings never reached the parent task. The developer is seeking advice on how to scope reviewer visibility and ensure output returns to the requesting agent.

This report highlights a critical flaw where review findings, unlike their cost and status, never reached the parent task, moving the problem from auditing to data flow.