dsh-jev-decide
DeepSeek Harness plugin registering a jev_decide tool so the agent can ask Noul, Choice or Score questions and gate on probabilities.
Overview
dsh-jev-decide connects TypeSafe's Jev to DeepSeek Harness by registering a single agent tool, jev_decide. When the agent needs a fast, accurate judgment, it calls Jev instead of letting the chat model guess. The tool takes a state, a question, a type (noul by default, choice or score), options or levels as needed, optional yes/no criteria to sharpen calibration, and a model name, and returns the answer with probabilities, confidence and usage.
The README, in Chinese, lists suitable uses (urgency triage, intent routing, guardrail checks, binary decisions, rubric scoring) and unsuitable ones (writing code or copy). Credentials resolve from plugin config, the TYPESAFE_API_KEY environment variable, or DSH's credentials file. It is published on npm and since v0.1.2 ships a bundle so dsh plugin add wires it automatically.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Yes / no | Any yes/no question about the state, with optional YES/NO criteria | yes · no |
| Choice | Pick one of the supplied options | — |
| Score | Rate on the supplied ordered levels | — |
How it works
- Agent calls jev_decide with state and question
- Plugin sends the request to Jev
- Agent receives answer, probabilities and confidence
- Agent gates its next step on the result
Strengths
- Simple single-tool integration
- Optional criteria to sharpen nouls
- Auto-wiring via bundle
Caveats
- Agent decides when to call it
- Chinese documentation
Getting started
dsh plugin --profile web add dsh-jev-decide, with TYPESAFE_API_KEY set.At a glance
- Type
- Plugin
- Status
- beta
- Language
- TypeScript
- Runtime
- DeepSeek Harness plugin
- Has code
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
Coding agents: DeepSeek Harness plugin whose single `jev_decide` tool lets the agent ask a `Noul`, `Choice`, or `Score` question about any state — urgency triage, intent routing, guardrail checks — and gate on the returned probability or confidence in code instead of trusting the chat model's guess.
yibie/awesome-jevThis DSH plugin registers TypeSafe Jev as an Agent tool that returns calibrated probabilities for noul, choice, and score judgments without generating text.
logicrw/awesome-jev-projectsDSH plugin: register TypeSafe Jev (System One decision model) as an agent tool…
wh000wh000/awesome-jev-liveDSH plugin: register TypeSafe Jev (System One decision model) as an agent tool — jev_decide returns calibrated probabilities (noul/choice/score) for routing/triage/guardrail judgments, no text generation. 把 TypeSafe Jev 决策模型注册为 DSH agent 工具.
hellogumbo/awesome-jev
Frequently asked questions
What does the tool return?
The answer, probability distribution, confidence and token usage.
What is it not for?
Writing code, copy or anything needing generated text.
Where does the key come from?
Plugin config, TYPESAFE_API_KEY, or DSH's credentials file.
Related projects
DeepSeek Harness Web GUI plugin giving the agent a jev_ask tool, usage tracking and an optional command guard with explicit caveats.
PluginbetaTypeScriptDeepSeek Harness bundle registering a jev_ask tool for typed noul, choice and score questions to Jev.
ProjectprototypeTypeScriptDeepSeek Harness plugin that gives the human a Jev second opinion beside the agent's multiple-choice questions.
ProjectbetaTypeScriptAsync Python intent and tool router: Jev picks the handler, and a code-side confidence gate stops weak or invented decisions.
ProjectbetaPythonTiny JavaScript semantic matcher that classifies text locally in microseconds and escalates ambiguous cases to Jev.
ProjectbetaJavaScriptTypeScript/Python policy engine where Jev only answers declared questions and code decides routes, gates and actions.
ProjectbetaTypeScript