AI Agent Hub
Back to plugins
⚙️

dsh-attention-beep

Workflow Updated 2026.09.15

Run the following command in DeepSeek Harness:

dsh plugin install kiterunner1/dsh-attention-beep

Paste the following prompt into your AI chat to install this plugin:

Run dsh plugin install kiterunner1/dsh-attention-beep in your terminal to install; the source is available at https://github.com/kiterunner1/dsh-attention-beep .

About this plugin

Running long commands through DeepSeek Harness on Windows comes with three recurring frustrations: you are on another tab and never notice when a task finishes; a stuck pwsh command silently consumes the entire turn; and the model keeps emitting needlessly expensive recursive enumerations that burn a minute for zero output. dsh-attention-beep exists to fix all three.

It does three things. First, audible alerts: seven event types (task end, tool warning, tool timeout, recovery, stall, question, approval) play through the host process via System.Media.SoundPlayer or WPF MediaPlayer, so the sound reaches you even if the browser tab is backgrounded, muted, or minimized. Voice assets ship as static files in the package, replacing the file changes the voice. Second, a stall watchdog: the criterion is not elapsed time but whether the DSH process subtree is still accumulating CPU and IO above an idle noise floor, sustained for a configurable silence window. When a stall is confirmed, only that single tool call is aborted and a recovery brief (what happened, shell state, how to proceed) is appended to the tool result so the model continues in the same turn. Third, a command guard: known performance pitfalls (Get-ChildItem -Recurse -Include, foreground Start-Sleep over 60 s, tilde passed to native commands, etc.) are rejected before spawn with a suggested rewrite, returning in roughly 0 ms.

Built for Windows users who run long pwsh workloads through DSH and want reliable feedback regardless of whether they are watching. The plugin makes no network requests and relies solely on OS-level facts (process CPU/IO), so stall detection applies to any long-running command without per-tool adaptation.

Use Cases

  • Audible alert plays on the host OS when a long task finishes, even if the browser tab is backgrounded or muted
  • A stuck pwsh call (waiting for input, deadlock, hung network) is auto-aborted and a recovery brief is appended so the model continues seamlessly
  • Known performance pitfalls like recursive Include or foreground long sleeps are rejected pre-spawn with a suggested rewrite, saving minutes

Best For

  • Developers running long PowerShell workloads through DeepSeek Harness on Windows
  • Users who need reliable task-state feedback even with the browser tab backgrounded or muted
  • Maintainers of automated workflows who want to prevent the model from repeatedly emitting high-cost commands