dsh-restart-autoresume

After a DSH restart, running top-level tasks simply stop and operators must manually re-submit their requests. dsh-restart-autoresume fills that gap: before the restart it captures eligible top-level sessions and flushes their durable state, delegates the actual restart to the existing dsh-restart Host service, and then safely submits one auditable continuation turn once the new Host is up, prompting the Agent to inspect external outcomes before retrying.

At the core of the plugin is a schema-versioned private Resume Ledger that provides serialized atomic updates, corruption isolation, expiry, capacity cleanup, and 0600 file permissions. Combined with stable intent provenance, leases, bounded concurrency, a retry ceiling, and manual re-arming at startup, both crash recovery and operator-initiated restarts follow the same predictable path. The plugin deliberately does not implement a second restart mechanism; it strictly depends on the dsh-restart 0.3.1 Host contract, and if the provider is absent or malformed the request fails fast before any Ledger state is written.

If your workflow runs on DSH 0.1.1-rc.2 with dsh-restart mounted and you want restarts to not interrupt running top-level Agent tasks, this plugin is worth trialing in a disposable or explicitly approved profile. It is in public alpha, does not restore subagents directly, and does not guarantee exactly-once semantics for arbitrary tools, but it provides a structured safety net for the most common keep-working-after-restart scenario.