Overview
@nethiel/dsh-reasoning-effort-sniffer
Sniffs which thinking levels each third-party pi-ai route actually accepts and declares them, so the composer's model picker offers a strength switch; a sidebar panel switches sniffing and strength per provider.
README / EN
Package documentation
Registry summary
Sniffs which thinking levels each third-party pi-ai route actually accepts and declares them, so the composer's model picker offers a strength switch; a sidebar panel switches sniffing and strength per provider.
dsh.pub verifies the pinned bundle contract, runtime facts, and distribution semantics. The complete README remains in the source repository.
Read the full README on GitHubLIMITATIONS
Known limitations
- 只处理 `openai-completions` 协议的手工路由；catalog 路由与非 OpenAI 协议不碰。 - 「端点接受参数」不等于「后端真的按档位改变思考长度」，实际效果取决于网关实现。 - 某些模型关不掉思考（例如 GLM 系列），这类模型不会声明 `off` 档。 - 关掉再打开「思考强度」且嗅探开着时，会重新探测该供应商的模型，受每轮预算限制可能分几轮完成。