dsh-plugin-jev-effort-selector
DeepSeek Harness plugin where Jev picks each message's reasoning level from the model's ladder, inheriting depth for follow-ups.
Overview
dsh-plugin-jev-effort-selector lets Jev decide how much reasoning each message deserves in DeepSeek Harness, where effort is otherwise switched by hand. Before the first model call of a turn it asks Jev once and rewrites that call's reasoning level, shown as a chip beside the model selector. Levels are derived from what each model advertises (lowest, medium, high), so models that cannot disable thinking never get off, and max or xhigh are never used automatically.
A fixed-size context envelope of about 500 to 800 tokens tells Jev the previous level, what the user and assistant said, how much work was done and whether it finished, without sending history. Jev is asked both the level and whether the message continues the previous task; the one hard rule is that continuing unfinished work never drops below the previous level. Low confidence breaks ties upward, manual choices are respected, subagents are left alone, and every failure silently keeps the existing level. Documentation is in Chinese with an English README.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which reasoning level does this message need? | — |
| Yes / no | Is this message continuing the previous task? | yes · no |
How it works
- Before the turn's first model call, build the context envelope
- Ask Jev the level and the continuation question
- Apply the hard rule and upward tie-breaking
- Rewrite the call's reasoning level; on failure keep the original
Strengths
- Avoids wasting high effort on greetings
- Follow-ups inherit topic depth
- Silent fallback on any failure
- Per-session toggle
Caveats
- Specific to DeepSeek Harness
- Chinese-first documentation
Getting started
dsh plugin --profile web add github:justhalfbit/dsh-plugin-jev-effort-selector, then restart dsh web.At a glance
- Type
- Plugin
- Status
- beta
- Language
- TypeScript
- Runtime
- DeepSeek Harness plugin
- Has code
- yes
Use cases and integrations
How curated lists describe it
DeepSeek Harness (DSH) 推理等级自动选择插件：由 Jev System One 模型判断每条消息值多少思考量，按模型声明的等级自动推导档位，上下文信封让「继续」这类追问继承话题深度，低置信度向上取，任何失败都静默沿用原等级。 | Jev-driven reasoning effort per message: per-model ladders derived from what each model advertises, a fixed-size context envelope so follow-ups inherit topic depth, ties break upward, silent fallback on every failure path.
hellogumbo/awesome-jev
Frequently asked questions
What does Jev decide?
The reasoning level for each message and whether it continues the previous task.
What happens on low confidence?
It picks the higher of the two most likely levels.
Does it override my manual choice?
No, manual changes in the selector are respected for that turn.
Related projects
DeepSeek Harness plugin that asks Jev how much reasoning each prompt needs and sets reasoningEffort from the model's own levels.
ProjectbetaTypeScriptLocal-first routing engine for long-running agent tasks that uses a Jev decider to pick a model group per turn.
PluginproductionRust, TypeScriptOpen-source AI gateway for 100+ LLMs that can use Jev to classify requests for complexity routing and to compact tool results.
ProxyproductionPythonOpen-source workspace for running and reviewing AI coding agents across devices, with an optional Jev-based automatic model router.
ProjectproductionTypeScript- jev-router★ 593
Per-turn model routing for Claude Code and Codex: wraps the real CLI and lets Jev send simple turns to fast models and hard ones to strong.
ProxybetaJavaScript - jev-codex-router★ 274
Per-call router for Codex where Jev picks the model and thinking effort for every model call, including tool continuations.
ProjectbetaPython