What should a reusable Claude handoff file contain before another assistant takes over?

Copying a full Claude conversation preserves too much noise, while a short summary often drops the decisions that matter. A reusable handoff file could separate the goal, current state, authoritative files and links, constraints, decisions with rationale, failed approaches, validation already completed, open questions, and the next concrete step. It should also distinguish facts from assumptions and note which outputs are safe to overwrite. For people who move work from Claude to another assistant or back again, which fields consistently prevent repeated work or lost context? Does a human-readable Markdown file hold up, or do you eventually need structured JSON alongside it?