dsh-auto-driving

Long-running agent tasks in DSH Desktop are fragile: a single provider hiccup (quota exhaustion, rate limiting, a 5xx) can kill the run mid-flight, and a pending permission prompt will stall everything until someone clicks approve. dsh-auto-driving gives your agent tasks the ability to self-heal, so they find their own way forward when things go wrong.

The plugin works in three layers. Model auto-fallback: when a request fails before producing any visible output, the plugin cycles through your configured provider groups by priority and re-dispatches the same request to the next model, letting the in-progress task continue seamlessly. Task auto-retry: if every candidate fails due to transient errors like network timeouts or connection drops, the plugin retries on the same model with exponential backoff, and a built-in watchdog detects and recovers silent hangs. Full-auto mode: when enabled, all permission approvals, user confirmation questions, and plan reviews are auto-answered as allowed, so the task never waits for a human; every auto-decision is appended to an AUTO-MODE.md audit file in the workspace root for full traceability. A one-click toggle in the chat input bar activates any of these instantly, with no restart needed.

Ideal for DSH Desktop users who have configured multiple model providers and want their agent tasks to run unattended for extended periods. If your workflow routinely hits provider quota swings, network blips, or you simply do not want to babysit a running task to click approve, this plugin lets the task carry on its own.