WeChat model gateway
sulfide2085/dsh-llm-wechat · skills
DeepSeek Harness WeChat gateway adapter plug-in: reuse DeepSeekAdapter + streaming think tag translation
★ 6GitHub stars
0Forks
2026-08-15Last updated
JavaScriptLanguage
—License
Key features
- Response sideInsert the ThinkTagSplitter state machine before translate (processing tags that are cut across chunks, think with angle brackets, think with unclosed pockets, etc.).
- With this plug-in installed, the wechat supplier's model will automatically have an inference level drop-down (off / high / max) in the model selector without any additional configuration.
- dsh.bundle.patch declares the bundle layer (this package comes with cordis.patch.yml)
- System enhancementWeChat model's compliance with "Tools can only be called through run_code" is weak, and the plug-in appends a strong constraint at the end of system (only valid for the wechat channel).
Requirements
- 60 seconds request timeoutThe max file is very long to think about (up to 19k+ characters in actual measurement), and it is easy to hit the gateway and cut off in 60 seconds → TIMEOUT / 408.
- It is recommended to use high gear for daily use
Install command
dsh plugin --profile web add github:sulfide2085/dsh-llm-wechat