dsh-commandcode

Using Command Code models inside DSH-Desktop has long been hampered by sluggish catalog refreshes, single-key rate-limiting, and vague error messages when streams drop. dsh-commandcode keeps full feature parity with the original community plugin while adding a layer of operational robustness: a Stale-While-Revalidate model catalog with ETag conditional requests and a circuit breaker serves cached results instantly and refreshes in the background; multi-account configuration lets the plugin rotate automatically on 429 or 401 responses and probe active windows to revive cooled-down credentials; per-request and stream-idle timeouts are adjustable from the settings page, with exponential backoff plus jitter on transient failures.

Streaming generation covers every event type in the Command Code Provider SSE/JSONL protocol—text deltas, reasoning phases, tool calls, and usage stats—while every error carries a stable code, diagnostic context (status, model, account, request ID), and a plain-language troubleshooting hint. The redesigned settings page is organized into Connection, Usage and Plan, Multi-Account, and Advanced cards, with real-time form validation, full Chinese/English localization that follows browser language, per-account usage tabs showing five-hour window and weekly quota progress bars, plan-aware model filtering, and image-attachment support for vision-capable models.

It is built for individual developers and small teams running DSH-Desktop 0.7.x who rely on Command Code models and need multi-account rate-limit resilience, transparent usage tracking, and a stable inference experience. No manual build step is required—add the plugin to your profile bundles and restart to get started.