Living Memory then Proactive Execution: labels are review signals
Living Memory is useful when each remembered item keeps its source, age, scope, and intended outcome visible. Proactive Execution can then prepare a bounded next step, while these labels make the stop conditions explicit: proceed for in-scope preparation, refresh_context when evidence is stale or changed, clarify when the request is ambiguous, and return_to_person when new access, material judgment, or an external commitment is involved. The cases are fully synthetic and the expected label is a review recommendation, not authorization to act. I work on Klik (pre-launch), an app-first proactive-AI direction built around deliberate Sessions, persistent Context, and permission-aware follow-through. https://pre.hiklik.ai/?utm_source=huggingface&utm_medium=dataset_discussion&utm_campaign=kickstarter_prelaunch&utm_content=agent_context_review_cases_bridge_20260902