应用产品

OpenSmoke

@aaravriyer1931PythonMIT更新于 2026-09-18可当场跑

查找因环境问题(缺少密钥、工具、文件、权限、网络或上下文)导致运行失败的 AI agent,基于 TypeSafe Jev。

英文原文

Find the AI agent runs that broke because their environment did: missing keys, tools, files, permissions, network, or context. Powered by TypeSafe Jev.

aaravriyer193/OpenSmoke

它在哪儿调用了 Jev

from typesafe_sdk import AsyncTypeSafeClient, TypeSafeError

src/opensmoke/judges.py:95

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/opensmoke/questions.py:55

  1. claims_success是/否

    The final response presents the task as successfully completed.

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: {
    claims_success: { type: 'boolean', instructions: 'The final response presents the task as successfully completed.' },
  },
});

同类的其他项目