dsh-opencodego
DeepSeek Harness plugin: a single provider route 'opencode-go-native' that serves every OpenCode Go (and OpenCode-style OpenAI-compatible gateway) model with the correct wire protocol per model — openai-completions / openai-responses / anthropic-messages — resolved from the models.dev per-model `provider.npm` recommendation and verified against the gateway. Capability facts are NOT assumed: a per-model SYNC asks the gateway which protocols answer, verifies the official reasoning contract (falling back to the full ladder only when there is none), and reads context/output/modalities from a bundled models.dev baseline. The model list is a plain membership list; nothing is enabled until the operator picks it. Credentials live in the host credential store. Dual-face: the host half registers the adapter, the settings section and a narrow diagnostics/discovery/sync HTTP surface; the browser half adds an OpenCode Go page with '获取可用模型' (pick-to-activate) and '信息同步' (per-model capability sync with live progress).
安装
npx -p @deepseek-ai/dsh dsh plugin --profile web add github:whoiszzj/dsh-opencodego说明文档
阅读完整 README ↗配置
设置页里的字段:
| 字段 | 说明 |
|---|---|
baseURL | 网关地址,默认 https://opencode.ai/zen/go/v1,要带 /v1 |
| API Key | 存在 DSH 的凭据存储里,不落明文 |
| 会话头 | 网关要求的路由头,默认开,一般不用动 |
每个模型还可以单独改协议 / 上下文 / 输出 / 输入模态 / 思考档位(展开那一行的 ▸),改完即生效。