checkpointing-an-iteration

To save AI conversation state before context compaction, use a checkpointing workflow that appends recent work, decisions, and artifacts to an existing iteration journal, preserving chronological order and ensuring important information survives potential context loss.

The best way to preserve context during long AI conversation iterations is to execute a deterministic checkpoint workflow that updates your iteration journal entry with incremental progress before pausing or taking risky actions.

To create a checkpoint and save progress in an iteration journal, run the checkpoint command to append new information to the active journal file located at autonomy/[goal-name]/iteration-NNNN-YYYY-MM-DD.md.

You should save iteration progress to prevent context loss during long conversations, before mid-iteration pauses, or immediately prior to executing risky actions where important context must be explicitly preserved.

Yes, you need an existing journal file to save conversation checkpoints because the checkpoint workflow relies on appending new information to a pre-existing journal while maintaining chronological order.

To verify your iteration journal checkpoint was saved correctly, locate the journal file at autonomy/[goal-name]/iteration-NNNN-YYYY-MM-DD.md and confirm the "Work Performed" section contains your latest entries.