checkpoint

<!-- Adapted from pedrohcgs/claude-code-my-workflow (MIT), https://github.com/pedrohcgs/claude-code-my-workflow --> # Checkpoint ## Purpose Write a state snapshot the next session can resume from in under a minute — yours on another machine, a collaborator's, or a fresh context after compaction. The session log answers *what happened and why*; the checkpoint answers *where am I and what is next*: facts, file pointers, and actions. Three mechanisms look interchangeable; only one of them travels. **Auto-memory** is machine-local and never syncs. **`/rewind`** undoes within the current session. A **checkpoint** is a file committed to the repo, so it reaches anyone who clones it. That is the entire reason to write one — and the reason to commit it once written. ## When to use - Before handing the repo to a collaborator — or to yourself on another machine, where auto-memory does not follow. - Before a long break, a model switch, or the end of a working day — and when the `PreCompact` hook warns that compaction is imminent and mid-plan context is about to be compressed. ## When not to use - For the narrative of the session — that is the session log (`.claude/rules/session-logging.md`, template `.claude/templates/session-log.md`). The checkpoint links to the latest log by path; it does not retell it. - As a substitute for committing. Uncommitted work is still lost work: commit the work, then checkpoint, then commit the checkpoint. See `.claude/rules/autonomous-work.md`. ## ...