技术壁垒会重新铸造起来,最终还是和别人不一样的人重新站上顶峰,有些人是顶峰掉下去的,有些人是新来的,还有些人掉下去了再也没上来。其实有没有AI,世界都是这样的,只是重新洗了一次牌,总要给后来者机会。
blog.chain0x0.com/blog/%E6%9…
仿佛让我回到了挖前端漏洞的时光 这些js代码总是这么有趣 奈何就是不值钱啊 github.com/52-HRTZ/security-…
已经被哥几个吃了一年的技术公开了哈哈哈
File upload restrictions aren't aalways enough to prevent stored XSS, especially when the storage service itself provides ways to override how uploaded files are served.
MinIO is a good example because some configurations honour the `response-content-type` query parameter which can change the response's `Content-Type` without modifying the stored object. An application might force every uploaded file to be served as `image/png`, but adding `?response-content-type=text/html` to the object's URL could make the browser interpret that same file as HTML instead.
Amazon S3 supports the same parameter although it requires a signed request. When hunting around public uploads, it's worth paying attention to these response overrides, whether the uploaded content can contain HTML/JS, and especially which origin serves the files. Getting JavaScript to execute is one thing, but getting it to execute under an application's trusted origin is where the impact gets much more interesting.
CourseCrack benchmark上线之后
我用荒废了很久的harness跑了一遍 意料之中只拿到3个flag
这个harness最开始帮我挖了很多漏洞
到后来加门禁、加限制、加契约
越加越复杂 居然越来越没效果 就直接没用过了
昨天晚上熬了一个通宵用最初的架构改成比第一版还简单的版本
这个版本拿了8个flag 并且同时更短
这样来看 在复杂任务中 注意力其实最重要的
在遍地大参数模型的当下和更大参数模型的以后 设计harness要相信模型自身的能力 不能困在之前写工具的思维里 把所有情况都考虑到写成死代码 反而适得其反
Less is more 牛逼!
















