checkpointlisted

# /checkpoint — save full session state for clean restart ## Why this exists Claude Code session state is *partly* persistent and *partly* not: | Survives a restart? | Where it lives | |---|---| | ✓ Files on disk | regular filesystem | | ✓ User memory | `~/.claude/projects/*/memory/` | | ✓ Conversation context | only via `claude --continue` / equivalent | | ✗ TaskList | session-scoped, dies on exit | | ✗ Cron jobs from CronCreate | session-scoped, dies on exit | | ✗ Background `Bash run_in_background` processes | session-scoped, dies on exit | | ✗ Loaded skill changes since startup | reloaded fresh on next launch | | ✗ Anything in `/tmp` not on tmpfs persistence | depends on system reboot vs Claude restart | `/checkpoint` is a single command that drains the *not-survives* column to disk so `/restore` (in the next session) can re-hydrate the parts that can be rehydrated and tell you about the parts that can't. ## Behavior summary When you (the agent) execute `/checkpoint`: 1. **Wait for a safe stopping point.** Don't checkpoint mid-tool-call. If a long-running background process is mid-flight (`Bash` with `run_in_background: true`), explicitly mention it to the user and ask whether to: a) wait for it to finish before checkpointing b) leave it running (it'll die at exit) and note the loss c) cancel it via TaskStop and then checkpoint Default: ask, don't decide silently. 2. **Capture the data** (see Steps below). 3. **Report back** with a clear "saf