Context Window Management

To prevent context loss in long-running AI sessions, monitor context usage and update a central progress file. This skill enforces compression thresholds and spawns fresh agents when usage exceeds defined limits, ensuring critical decisions are preserved.

You should compress context when usage crosses 50% and spawn a fresh agent when usage exceeds 80%. This prevents context overflow during multi-phase workflows by enforcing strict thresholds and updating the state file accordingly.

Agent handoff works by updating the central progress file at .workflow/state/progress.md to preserve critical decisions. Fresh agents read this checkpoint to resume long-running sessions seamlessly without losing prior context.

Yes, context window management supports per-worktree isolation and explicit checkpoints. This enables seamless handoffs and robust recovery by isolating session state and tracking progress independently across worktrees.

Phase 4.5B Context Sync is a process for managing context windows to prevent information loss. It is needed during long multi-phase workflows and agent handoffs to enforce compression thresholds and trigger fresh agent spawning.