dsh-auto-update

Upgrading DeepSeek Harness on Windows has a hard blocker: native Node addons such as sharp, koffi, and node-pty keep their DLL handles locked by the operating system, so any hot-overwrite attempt fails with an EBUSY error. In containerised deployments on Docker, Kubernetes, or Podman, a naive process.exit call simply tears down the container before the new files ever land on disk. dsh-auto-update exists to eliminate exactly these two failure modes so that upgrades can complete cleanly regardless of the host environment. The plugin is built around a three-stage pipeline: stage, detached apply, and container-aware protection. While Harness is running, incoming updates are written into an isolated staging directory and the locked binaries are never touched. Once the main process exits and the OS releases every native handle, a fully detached child process performs an atomic file replacement. A per-install apply.lock guard prevents concurrent update processes from corrupting the global state, and both the Windows shim and the POSIX bin/dsh symlink are followed correctly so the next launch picks up the new version on any platform. In container contexts the plugin detects Docker, Podman, and Kubernetes markers before taking any destructive action. Instead of calling process.exit(0) and killing the container, it surfaces a clear message in the Web UI telling the operator to pull the latest image, leaving the actual rollout to the orchestrator. On the web dashboard a small collapsible capsule in the lower-right corner lets you compare versions, trigger a manual update check, or restart with a single click, without digging through terminal logs. If you run DeepSeek Harness on a Windows desktop, a Linux or macOS server, or inside a container fleet and want upgrades to feel as frictionless as any ordinary application update, dsh-auto-update is the piece that makes it happen.