Plan recovery of an interrupted AI agent run from checkpoints and action receipts; identify what can resume and which effects need checking before retry.
Use pasted traces, logs, configs, transcripts, checkpoints, files and action receipts to reconstruct what completed, prepare a safe resume plan, and check the result after an authorized restart. Distinguishes confirmed effects from uncertain attempts and keeps pending approvals explicit.
3 in the Claude listing