Pliny开源V3R1T4S：让Claude Code夜间自动破解未解难题
Pliny just open-sourced V3R1T4S: paste one line into Claude Code, and while you…
通过具体的“一夜破译两题”的真实案例和数据（135个题库、三重验证机制）来展示AI Agent的能力，结构清晰，有明确的行动指引和GitHub链接，写法值得参考。
Pliny just open-sourced V3R1T4S: paste one line into Claude Code, and while you sleep it spends 𝘆𝗼𝘂𝗿 𝗹𝗲𝗳𝘁𝗼𝘃𝗲𝗿 𝘂𝘀𝗮𝗴𝗲 𝗳𝗼𝗿 𝘁𝗵𝗲 𝘄𝗲𝗲𝗸 on problems nobody has solved.
Pliny 刚刚开源了 V3R1T4S：将一行代码粘贴到 Claude Code 中，在你睡觉的时候，它会消耗你本周剩余的用量去解决那些无人攻克的问题。
All Pliny sent Opus 5.5 was one sentence full of typos. By the next morning, it had cracked 𝘁𝘄𝗼 𝗽𝘂𝘇𝘇𝗹𝗲𝘀 𝗲𝘃𝗲𝗻 𝗰𝗶𝗽𝗵𝗲𝗿 𝗲𝘅𝗽𝗲𝗿𝘁𝘀 𝗵𝗮𝗱𝗻'𝘁 𝘀𝗼𝗹𝘃𝗲𝗱, and the puzzle setter personally added Pliny's name to the solver list.
Pliny 发给 Opus 5.5 的只是一句充满拼写错误的句子。到了第二天早上，它已经破解了两个连密码专家都未曾解决的谜题，谜题设置者亲自将 Pliny 的名字添加到了解题者名单中。
Paste this, answer 6 questions, and it picks a problem that matches your interests and keeps at it every night: set up V3R1T4S: https://github.com/elder-plinius/V3R1T4S
粘贴这段代码，回答 6 个问题，它就会挑选一个符合你兴趣的问题并每晚持续攻关： 搭建 V3R1T4S：https://github.com/elder-plinius/V3R1T4S
You can also name a target, like the German Army Enigma messages from 1941 that are still unbroken, or Linear A, which nobody can read yet. Its problem bank has 135 of them.
你也可以指定目标，比如至今未被破译的 1941 年德国军队恩尼格玛（Enigma）电报，或者至今仍无人能解读的线性文字 A。它的题库中包含 135 个此类难题。
Every time it claims a solve, 3 agents that never saw the original run each write their own code to reproduce it from scratch. If any one of them rejects it, it doesn't count.
每当它声称取得突破时，三个从未见过原始信息的代理会各自编写代码，从头开始复现结果。如果其中任何一个代理拒绝认可，该成果就不算数。
Nothing gets scheduled or published without your OK, and you set how long each run lasts. Most nights it finds nothing. But solve one, and your name goes in the trophy case.
未经你的许可，任何内容都不会被安排或发布，且你可以设定每次运行的持续时间。大多数夜晚它一无所获。但一旦解出一道题，你的名字就会进入荣誉榜。
更进一步：量化金融体系
看懂新闻只是起点——沿量化金融路径，把它变成能交付的工程能力