Run a second agent whose only job is deciding when to interrupt the first with a reminder (+8.3pp on Terminal-Bench 2.0)

arXiv 2607.08716·high signal

"Remember When It Matters" (arXiv:2607.08716, July 9, 2026) attacks behavioral state decay — where task-critical instructions get buried or pushed out of the context window on long-horizon runs — with a separate memory agent that maintains a structured bank and proactively injects task-relevant reminders into the action agent's context. It gained +8.3 percentage points pass@1 on Terminal-Bench 2.0 and +6.8pp on τ²-Bench, and helped both weak and strong action agents. Critically, selective intervention beat both passive bank exposure and always-on injection — so the value is in the *when*, not the retrieval, and it drops in as a plug-and-play wrapper around an existing framework.