dsh-llm-xai-oauth

If you already subscribe to SuperGrok or X Premium and do not want to pay separate xAI API metering, this plugin lets DeepSeek Harness use your subscription as a native LLM channel end to end: OAuth device-flow login, on-disk token storage, and on-demand refresh — all without touching the billable api.x.ai endpoints. It also solves a subtle but common problem: SuperGrok access tokens live only about an hour, and the dsh process itself does not refresh them offline. The plugin ships an independent daemon mode that re-reads the token roughly once a minute and refreshes five minutes before expiry, so you are not stuck with cascading 401s on /usage and chat after a laptop sleeps overnight. On the feature side, the plugin registers a live SuperGrok adapter route inside Harness, pulls the remote model catalog at startup (currently grok-4.6 with xhigh reasoning, grok-4.5, and grok-4.3), supports 500K to 1M context windows, and proactively raises CONTEXT_WINDOW_EXCEEDED when the remaining window cannot hold the requested max_tokens — handing off to Harness overflow compaction instead of letting a long conversation die with a generic 400. Every proxy, client ID, and endpoint is overridable via environment variables, and settings hot-reload without restarting dsh. It is aimed at developers who already run DSH with dsh-ssh-tui, hold a SuperGrok or X Premium subscription, and want Grok models in daily chat and long-form reasoning — particularly on remote SSH hosts or CI boxes where token expiry is a recurring headache.