What problem does it solve?
Prevents context overload, data loss, and runaway session state by providing a repeatable protocol to save, compact, and manage AI session memory before switching contexts or spawning background work.
Core Features & Use Cases
- Save & Persist: Save completed, in-progress, and pending work to dated memory files (memory/YYYY-MM-DD.md) and update MEMORY.md when significant.
- Compact Protocol: Run session_status checks, switch models for compacting (sonnet → opus), and confirm readiness before and after compaction to preserve continuity.
- Sub-agent & Async Safety: Schedule one-shot check-back crons before spawning sub-agents or promising timed work, ensure main session owns follow-ups, and enforce fail-fast behavior for sub-agents near context limits.
- Context Thresholds: Warn at 50% and take automated actions at 75% token usage with different behaviors for main sessions versus sub-agents.
Quick Start
Save the current session to memory/YYYY-MM-DD.md, schedule a check-back cron for the task, then spawn the sub-agent so the main session retains follow-up control.