AI Decision System
Replayable, offline-verifiable records for AI-assisted go/no-go decisions: experiment promotions, typed model answers and human-approved actions.
翻译准备中
尚未检测到更新。GitHub 发布版本后会显示在这里。
个人 GitHub 账号下公开仓库的所有者可以登录并验证所有权,自行管理介绍,无需管理员批准。
登录并生成验证文件,在 30 分钟内将其提交到此仓库的默认分支。其他分支或分叉仓库中的文件无效。然后在下方检查文件,即可无需管理员批准直接管理项目。请保留文件以维持管理权限。
如需登录,将前往登录页面。登录后请回到此处,再次点击所选的验证按钮。
在项目文件夹中运行 /nomorevibe verify,即可验证所有权并自行管理介绍。
一分钟完成验证A crash test for AI agents: what does your agent tell you when its tools lie or fail?
翻译准备中
取自仓库 README
正在加载完整 README。 以下为已保存的摘录。
A crash test for AI agents: what does your agent tell you when its tools lie or fail?
Ajantik puts an agent in front of a tool layer that misbehaves on purpose: a save that
reports success and stores nothing, a reply cut off mid-JSON, a record that arrives empty,
a 503. It records every call together with the real state of the world after it, then
judges two things separately:查看完整 README ↗目前无法加载完整 README。请通过原文链接阅读。 以下为已保存的摘录。
A crash test for AI agents: what does your agent tell you when its tools lie or fail?
Ajantik puts an agent in front of a tool layer that misbehaves on purpose: a save that
reports success and stores nothing, a reply cut off mid-JSON, a record that arrives empty,
a 503. It records every call together with the real state of the world after it, then
judges two things separately:查看完整 README ↗