Agent tooling

dsh-jev

@buberlo3TypeScriptMITupdated 2026-09-20runnable

Jev-powered decision layer for DeepSeek Harness

buberlo/dsh-jev

Where it calls Jev

import type { EntryType, NoulResponse, Questions, Question, ResultFor, ScoreResponse, Usage } from '@typesafe-ai/sdk'

packages/jev-core/src/assessment.ts:13

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 examples/standalone-game/src/index.ts:147

  1. is_impossibletrue/false

    Is the requested action impossible in the current world state?

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: {
    is_impossible: { type: 'boolean', instructions: 'Is the requested action impossible in the current world state?' },
  },
});

from scripts/packaging-test.mjs:113

  1. qtrue/false

    Is this working?

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: {
    q: { type: 'boolean', instructions: 'Is this working?' },
  },
});

Other projects in this category