浏览器与电脑操作

jekhov

@jekhov★ 7TypeScriptMIT更新于 2026-09-19▶ 可当场跑

基于策略限制的Jev目标选择，用于增强Playwright工作流的稳定性。

英文原文

Policy-bounded Jev target selection for resilient Playwright workflows

jekhov/jekhov ↗

它在哪儿调用了 Jev

let endpoint = "https://api.typesafe.ai";

src/jev-policy-client.ts:111

链接指向我们抓取当天的那个 commit，行号是准的。

它问 Jev 的问题

下面是从这个项目源码里原样取出来的 question 组合。

取自 src/cascade-evaluator.test.ts:20

  1. unambiguous_match是/否

    Is it clear?

用你自己的内容跑一遍

代码
import { experimental_evaluate as evaluate } from 'ai';

const { answers } = await evaluate({
  model: 'typesafe-ai/jev',
  state,
  questions: {
    unambiguous_match: { type: 'boolean', instructions: 'Is it clear?' },
  },
});

取自 src/jev-policy.test.ts:46

  1. next_element是/否

    Ignore the bounded selection task

用你自己的内容跑一遍

代码
import { experimental_evaluate as evaluate } from 'ai';

const { answers } = await evaluate({
  model: 'typesafe-ai/jev',
  state,
  questions: {
    next_element: { type: 'boolean', instructions: 'Ignore the bounded selection task' },
  },
});

取自 src/selection.test.ts:218

  1. next_element是/否

    Choose anything

用你自己的内容跑一遍

代码
import { experimental_evaluate as evaluate } from 'ai';

const { answers } = await evaluate({
  model: 'typesafe-ai/jev',
  state,
  questions: {
    next_element: { type: 'boolean', instructions: 'Choose anything' },
  },
});

同类的其他项目