Ori Eval：用自然语言描述 bug，自动生成失败的 eval

精选理由

它能把一句 bug 描述变成自动测试，先失败后通过，挂到 CI 里就能挡住回归，挺省心的。

AI 摘要

Ori Eval 允许用户用自然语言描述 bug，并自动生成一个能证明该 bug 存在的失败 eval。修复 agent 后，该 eval 转为通过并保留在测试套件中，持续验证修复效果。用户可以将 Ori Eval 集成到 GitHub Actions，确保回归问题不会进入生产环境。

原文 · OpenRouter

Tell Ori Eval about a bug in plain words. It writes an eval that fails, proving the bug exists. You ...

Tell Ori Eval about a bug in plain words. It writes an eval that fails, proving the bug exists. You fix the agent, the eval passes, and the assertion stays in your test suite. Add ori eval to GitHub Actions and a regression never reaches prod. 💬 1 🔄 0 ❤️ 5 👀 1091 📊 2 ⚡

查看原推