dsh-web-restart

One-click restart button for the DeepSeek Harness Web UI.

Adds a small circular restart button (↻) to the sidebar footer, next to Settings. A single click immediately restarts the dsh web process — the page disconnects for ~15–20 seconds, then you refresh and everything is back (sessions are persisted on disk and recover automatically). The button is persistent: it survives the restart it triggers.

Windows-specific: the restart launches dsh web through an independent hidden PowerShell process, so the script keeps running after the harness process is killed.

Features

• Single-click restart — no confirmation step, no double-click dance.
• Persistent: survives the DSH restart it triggers (installed as a bundle-layer plugin, not a dynamic session plugin).
• Small footprint: sits beside the Settings trigger in the sidebar footer; icon-only in the 56px rail, icon + label in the wide sidebar.
• In-flight feedback: the button turns red and shows "重启中…" → "已触发" while the request is being processed.
• Re-entry guard: a second click while a restart is already in flight is rejected.
• Online status dot: the button also shows DSH liveness — a green dot polls GET /dsh-health every 5s (red when the harness is unreachable/restarting).

Install

From GitHub (this repo)

dsh plugin --profile web add github:YOUR_OWNER/dsh-web-restart

Then restart dsh web once so the bundle layer loads.

Manual (edit profile files)

1. Add the dependency:

// ~/.dsh/profiles/web/package.json
{
  "dependencies": {
    "dsh-web-restart": "github:YOUR_OWNER/dsh-web-restart"
  }
}

1. Add the plugin row (this repo ships its own cordis.patch.yml — either merge its row into your profile's cordis.patch.yml, or install via the dsh plugin command above which does it for you):

# ~/.dsh/profiles/web/cordis.patch.yml
- insert:
    - id: dsh-web-restart
      name: dsh-web-restart

1. pnpm install in the profile directory, then restart dsh web.

How it works

The host routes return before the restart happens (~1s host timer + 3s inner sleep), giving the browser time to render the "restarting" state before the page drops.

Why not ctx.timeout in the route handler

ctx.timeout is a Cordis mixin backed by ctx.effect(), which is scoped to the Cordis fiber lifecycle. A webServer route handler runs as a plain Node HTTP callback — outside any fiber — so ctx.effect-registered timers are silently dropped and the restart never fires. This plugin uses Node's native setTimeout in the handler instead.

Compatibility

• DeepSeek Harness 0.1.0-rc.6 (web profile).
• The restart script path (D:\1\dsh-web-host.ps1) is hard-coded for this author's machine — fork and adjust it (or replace the command in lib/index.js) for your own environment. Contributions to make this configurable are welcome.

Development

The client bundle is hand-written in the exact wire format (window.__ModuleLoader__.load({ id, factory })), so no build step is required:

node --check lib/index.js
node --check lib/client.js

License

MIT