Agent tooling

jev-dsh-decision

@Devin-AXIS4JavaScriptupdated 2026-09-20runnable

Jev DSH 决策引擎|面向 Agent Harness 的结构化决策插件。原生支持 DeepSeek Harness,通过 iPolloWork 支持 OpenCode、Codex Harness。

Devin-AXIS/jev-dsh-decision

Where it calls Jev

const ENDPOINT = 'https://api.typesafe.ai/v1/systemone';

service/jev.mjs:4

The link points at the commit we read, so the line number still holds.

What it asks Jev

These question sets are lifted from this project's source exactly as written.

from service/jev.mjs:147

  1. connectedtrue/false

    Does this text describe a connection check?

Run it on your own text

The code
import { experimental_evaluate as evaluate } from 'ai';

const { answers } = await evaluate({
  model: 'typesafe-ai/jev',
  state,
  questions: {
    connected: { type: 'boolean', instructions: 'Does this text describe a connection check?' },
  },
});

from tests/jev.test.mjs:6

  1. ownerchoice

    Which agent?

    • writerWrites text
    • slidesCreates slides
  2. qualityscore

    Rate clarity.

    • 0unclear
    • 1clear
    • 2excellent
  3. reviewtrue/false

    Need a reviewer?

Run it on your own text

The code
import { experimental_evaluate as evaluate } from 'ai';

const { answers } = await evaluate({
  model: 'typesafe-ai/jev',
  state,
  questions: {
    owner: { type: 'choice', instructions: 'Which agent?',
      criteria: { writer: 'Writes text', slides: 'Creates slides' } },
    quality: { type: 'score', instructions: 'Rate clarity.',
      criteria: ['unclear', 'clear', 'excellent'] },
    review: { type: 'boolean', instructions: 'Need a reviewer?' },
  },
});

from tests/work.test.ts:64

  1. need_slidestrue/false

    Does this task need slides?

Run it on your own text

The code
import { experimental_evaluate as evaluate } from 'ai';

const { answers } = await evaluate({
  model: 'typesafe-ai/jev',
  state,
  questions: {
    need_slides: { type: 'boolean', instructions: 'Does this task need slides?' },
  },
});

Other projects in this category