Introduction¶
When integrating an external LLM route in DeepSeek Harness (DSH), it is necessary to first verify that text output, streaming output, tool calling, credential configuration, and error handling are available. dsh-plugin-commandcode-provider adds a commandcode LLM provider route to DSH, allowing models to be selected in the form of commandcode/<model-id>.
What is this¶
- 定位: Adds a
commandcodeLLM provider route to DSH. - Repository:
mitian233/dsh-plugin-commandcode-provider. - License: MIT.
- Compatible Version: DSH
0.1.0-rc.6and later versions (npmnexttag). - Source: The browser-assisted OAuth flow is ported from the MIT-licensed
pi-commandcode-provider.
Core Capabilities¶
The verified capabilities are introduced below:
- Supports text messages and streaming text output.
- Supports standard JSON Schema tools and full tool calling.
- Supports
reasoning deltaswithin the reasoning intensity declared by the model. - Supports usage reporting, cancellation, idle timeout handling, and deterministic HTTP/stream error classification.
- Provides a Command Code OAuth button in the Web settings page for browser-assisted API key retrieval and storage.
- Supports overriding
baseURL,maxTokens,defaultContextWindow,streamIdleTimeoutMs, etc., via DSH profile configuration.
Installation and Usage¶
First, run the web version of DSH:
npx @deepseek-ai/dsh web
If dsh is already in PATH, you can also use dsh web.
Install from a local directory to the web profile:
cd /path/to/dsh-plugin-commandcode-provider
npx @deepseek-ai/dsh plugin --profile web add .
Install from GitHub:
dsh plugin --profile web add github:mitian233/dsh-plugin-commandcode-provider
Remove the plugin and its bundle layer:
dsh plugin --profile web remove dsh-plugin-commandcode-provider
For verification, first check the plugin layer, then start the Web UI:
dsh --profile web --dump-config
dsh web
Typical Usage¶
Add a plugin entry named llm-commandcode to the profile configuration:
- id: llm-commandcode
config:
apiKeyEnv: COMMANDCODE_API_KEY
baseURL: https://api.commandcode.ai
maxTokens: 64000
Use commandcode/<model-id> when selecting a model, for example:
commandcode/gpt-5.6-luna
The API key is configured via the DSH credentials service. If a credentials service is mounted, it is the authoritative source; otherwise, it will not fall back to the launch environment. Keys that are blank, whitespace-only, or HTTP-header-invalid will be rejected before sending the request.
The Command Code OAuth button in the Web settings page can initiate a browser-assisted flow to retrieve and store the API key. This flow requires the DSH Web server (Web profiles). If the automatic transfer fails or times out, you can fall back to the manual API key input below the button.
Configuration Options¶
apiKeyEnv: Used to select the DSH credential reference, e.g.,COMMANDCODE_API_KEY.baseURL: Used to modify the API origin.maxTokens: The upper limit is64000.defaultContextWindow,streamIdleTimeoutMs: Used to override model context window and streaming idle timeout behaviors, etc.temperature: Accepted as a compatibility configuration; explicit requests take precedence, with the wire default value being0.3.- Static
modelsentries: Can provide display, context-window, and max token overrides; it is not used for model discovery.
Use Cases and Notes¶
This plugin is suitable for scenarios where you need to use the commandcode provider route in DSH and manage the API key via the Web settings page.
The following are v1 non-goals:
- Does not refresh credentials.
- Does not perform dynamic model discovery or directory caching.
- Does not support image input and
stopsequences. - Image and
stoprequests will fail before network distribution.
The plugin runs with the permissions of the current dsh process and handles API key-related configurations. You should check the source code and license before installing.
Links¶
- Directory Page: https://www.skillhub.cn/plugins/mitian233/dsh-plugin-commandcode-provider
- GitHub: https://github.com/mitian233/dsh-plugin-commandcode-provider