dsh-zen-spoof

OpenCode Zen gates its free tier behind HTTP header checks that verify requests originate from the official OpenCode client. The pi-ai adapter bundled in DeepSeek Harness has a subtle flaw: its fixed attribution headers silently overwrite any user-supplied User-Agent, so the spoof never actually takes effect. A local installer patch can fix this temporarily, but it is not part of the npm package and can disappear after a dsh upgrade, leaving the whole setup fragile.

dsh-zen-spoof takes a different route: it registers as a standalone opencode provider, builds every spoof header per request, and issues the fetch directly to the Zen gateway, completely bypassing pi-ai's header pipeline. The plugin is fully self-contained, depends on no patch, and works with any dsh version. On startup it pulls the latest free-model list from the Zen API, so the catalogue stays current without manual upkeep.

On the streaming side, the plugin maps reasoning_content into reasoning blocks and content into text blocks, aligning with dsh's BlockAssembler protocol so the model's chain of thought appears natively in the UI. When Zen returns 429 or 5xx responses, it rotates through its candidate pool with exponential backoff and Retry-After support. API keys are resolved per request, so swapping credentials requires no restart. If you want cost-free access to OpenCode Zen models inside DeepSeek Harness without relying on a brittle local patch, this is the cleanest option available today.