[GGzili]

…ning

• Add restore <backup> command to restore SQLite from .bak backup files
• Add prune-backups --keep N command to remove old backups, keep N most recent
• Add --dry-run flag to sync command for preview without writing
• Add 50-session desktop cap warning to status command
• Full i18n support (EN/ZH_HANS) for all new features
• Auto-list available backups when no path argument given to restore
• Backup directory auto-discovery alongside SQLite file

[Wangnov]

Thanks for adding these maintenance features. The direction is useful, but I think this PR needs a few changes before merge, mainly around data-safety semantics.

A few notes based on the current main: #7 has already landed with the Windows hidden VBS startup path and follow-up fixes, so please rebase this branch onto current main and rerun the test suite before final review. This PR only appears to touch src/main.rs, so it should not need to modify the Windows service path.

Blocking items:

1. restore without an explicit backup path should not restore anything

Right now codex-threadripper restore lists available backups, picks the first backup, and then proceeds to restore it unless --dry-run is set. That is too dangerous for a destructive command. A user running restore with no args is likely expecting a list/help-like action, not an implicit overwrite of state_5.sqlite.

Suggested behavior:

• restore with no backup arg: list available backups and exit without writing
• restore <BACKUP_PATH>: perform the restore
• if we want “restore latest”, make that explicit, e.g. restore --latest, ideally with a confirmation/force flag later

2. prune-backups should not rely on filesystem creation time for ordering

The backup filenames already include a timestamp. Please sort backups by the timestamp parsed from the filename instead of metadata().created(). Filesystem birth/creation time can be unavailable or misleading after copies, restores, downloads, or archive extraction; using it to decide which backups to delete can remove the wrong files.

Suggested ordering:

• parse state_5.sqlite.<timestamp>.bak
• sort by parsed timestamp descending
• for unparseable .bak files, either skip them from automatic pruning or put them at the end with a warning

3. sync --dry-run currently only previews SQLite changes

The non-dry-run sync path can also update rollout JSONL metadata, but the dry-run path only reports SQLite distribution/mismatched rows. That makes the dry-run output incomplete and potentially misleading.

Please either:

• add a rollout dry-run scan that reports would-update / would-prepare / would-skip counts without writing, or
• clearly label the current output as SQLite-only dry-run

4. Please update README/help coverage for the new commands

The new restore, prune-backups, and sync --dry-run behavior should be documented, especially because restore/prune are data-affecting operations.

Once these are fixed, I think the feature set itself is worth having.