preserving-context

To preserve context and resume work, you create a structured checkpoint before compaction occurs. This checkpoint captures your current task, file path, progress, completed items, and remaining steps, enabling seamless resumption once the session context is lost.

Context checkpointing is the process of proactively saving your task state, including current file, progress, decisions, blockers, and test state. It prevents data loss during context switches across long-running, multi-file software engineering sessions.

You should create a checkpoint proactively before context compaction or task switches. Waiting until after compaction means the context is already gone, so you must capture the state beforehand to resume seamlessly later.

A task checkpoint schema should include the current task, file path, progress, completed items, remaining steps, decisions, blockers, and test state. This structured data ensures the agent can reload the full context.

Yes, you can use checkpointing to track task status by integrating with task-tracking tools. This aligns your task status updates with the preserved context, ensuring progress and remaining steps stay synchronized.