Skip to content

v0.32.0

Choose a tag to compare

@hhff hhff released this 10 Nov 18:05
· 2987 commits to main since this release
96c1eb4

What

  • recover (#180): brings a repo's supervisor back after a restart, on a laptop or the box, in tmux or in a terminal. It resumes the exact conversation where the harness can, which today means Claude Code; others start fresh with their opening prompt. recover --machine brings back every project on the machine.
    • It leaves alone a supervisor that's already running, a parked or drained fleet, and one the owner closed. It never starts two supervisors on one repo, it's safe to run twice, and --dry-run shows what it would do.
    • To know what was running before the machine went down, each dashboard's tick and the supervisor's own commands write run/supervisor-up.json, plus an entry under ~/.devteam/repos-<hub port>/. Nothing is read from the log.
    • An unattended restore (the box) and --force use the owner's resume_command as configured.
  • cloud/box-supervisors: the box's restore now calls recover for repos whose kit has it, and keeps the old path for older kits.
  • Docs: supervisor.md (recovery, and what the records can and can't prove), operator.md, README.

Why

The owner's requirement: sessions must survive a dead machine, a laptop battery or a cloud restart, whether in tmux or a terminal. Until now, only the box could bring supervisors back after a restart.

Migration

Restart each dashboard, so its tick starts writing the record recover reads. After a power cut on a laptop, run node docs/devteam/scripts/devteam.mjs recover --machine from any adopted repo. This release changes the box image, so the box restarts. Its repos keep the old restore path until they adopt this release.