Yasu 安利

#opensource #linux #security Knowledge is power, The Art of War..

[Yasu 安利 @ripmeout]

Retweet deleted tweet | https://t.co/hCRjEb3y1K

[Michael Guo @Michaelzsguo]

Lauren Tan @poteto 是 Cursor 的工程师，之前在 Meta 做 React Compiler，也在 Netflix 做过 tech lead 和工程经理。

她加入 Cursor 只有五个月。第一个月还在熟悉代码库，上个月已经合入了 1000 个 PR。这个月才过去 12 天，她又合入了接近 800 个。

这不是 AI slop code，而是你每天都在使用的 Cursor 的代码。

很多人，包括 Claude Code 的 Boris，都提过自己借助 coding agent 达到了类似的效率。但真正愿意把工作方法完整分享出来的人并不多。Lauren 在这个一小时的视频里，几乎是手把手讲了她怎么走到这一步。

她认为，用 AI coding 最大的问题不是生成代码，而是验证代码。

如果 agent 不能自己运行产品、操作界面、读取 CPU trace 和 heap snapshot、打开模拟器并复现问题，那么最后还是要由你来检查结果。你就是整个流程的 verifier，也是无法并行工作的瓶颈。

Lauren 的做法，是先给 agent 建立完整的验证能力：让它能通过 Chrome DevTools 或模拟器实际操作产品，再用 feature map 告诉它每个功能在哪里、怎么进入。这样即使同事只丢来一张截图，或者一句很模糊的 bug 描述，agent 也能找到对应功能，复现问题并验证修复。

每当她发现 agent 在猜测、漏读代码或走错方向，就把这个失败模式写成一条 skill。然后像测试代码一样测试这些 skill：让多个 sub-agent 分别执行任务，由 coordinator 制定 rubric，再让另一个模型交叉检查评分，反复迭代到结果足够稳定。

现在，她甚至允许 agent 自动合并 PR。有一天早上醒来，已经有 20 个 PR 自动进入 main；她直接在 main 上检查，结果都没有问题。

这套方法不是简单的提示词技巧，更像做工程管理：先设计好环境、流程和验收机制，再让团队并行工作。只不过这支团队，现在由几十个 coding agent 组成。

[Yasu 安利 @ripmeout]

w00t!

[科技混子Chris @ChrisSlacker]

有人发布了一个完全越狱、开源、预构建的 iOS 26 虚拟机

仓库: https://t.co/X5N1tfJuBB

[East @East_ML]

This is actually pretty terrifying 🤣

With Angels dead, Shakkou finds a new home and alongside Minerva and Merva, he seems like a great fit.

[Nitin Gavhane @NitinGavhane_]

📂 BUG BOUNTY
┃
┣ 📂 Recon
┃ ┣ 📂 Subdomain Enumeration
┃ ┣ 📂 Asset Discovery
┃ ┣ 📂 Port Scanning
┃ ┣ 📂 HTTP Probing
┃ ┣ 📂 DNS Enumeration
┃ ┗ 📂 Tech Detection
┃
┣ 📂 Web
┃ ┣ 📂 IDOR
┃ ┣ 📂 XSS
┃ ┣ 📂 SSRF
┃ ┣ 📂 SQLi
┃ ┣ 📂 CSRF
┃ ┣ 📂 SSTI
┃ ┗ 📂 File Upload
┃
┣ 📂 API
┃ ┣ 📂 API Discovery
┃ ┣ 📂 BOLA
┃ ┣ 📂 Authentication
┃ ┣ 📂 Authorization
┃ ┣ 📂 Rate Limiting
┃ ┗ 📂 Mass Assignment
┃
┣ 📂 Auth
┃ ┣ 📂 OAuth
┃ ┣ 📂 JWT
┃ ┣ 📂 Sessions
┃ ┣ 📂 MFA
┃ ┗ 📂 Account Takeover
┃
┣ 📂 Automation
┃ ┣ 📂 Recon Automation
┃ ┣ 📂 Nuclei
┃ ┣ 📂 Custom Scripts
┃ ┣ 📂 Asset Monitoring
┃ ┗ 📂 Workflow Orchestration
┃
┣ 📂 Burp Suite
┃ ┣ 📂 Proxy
┃ ┣ 📂 Repeater
┃ ┣ 📂 Intruder
┃ ┣ 📂 Extensions
┃ ┗ 📂 Match & Replace
┃
┣ 📂 Cloud
┃ ┣ 📂 AWS
┃ ┣ 📂 Azure
┃ ┣ 📂 GCP
┃ ┣ 📂 S3
┃ ┗ 📂 Misconfigurations
┃
┣ 📂 Reporting
┃ ┣ 📂 Validation
┃ ┣ 📂 PoC
┃ ┣ 📂 Impact
┃ ┣ 📂 Severity
┃ ┗ 📂 Report Writing
┃
┗ 📂 Growth
┣ 📂 Writeups
┣ 📂 CVE Research
┣ 📂 New Attack Surfaces
┣ 📂 Workflow Optimization
┗ 📂 Continuous Learning

The goal isn't to use more tools.

It's to build a workflow where every step feeds the next.

Recon → Discover → Test → Validate → Report

That's bug bounty orchestration.

#BugBounty #CyberSecurity #AppSec

[Yasu 安利 @ripmeout]

Park for later…

[SunSec @1nf0s3cpt]

A Security Researcher Reworked OpenAI’s CDC Prompt — and Found a $500K RCE for About $25 in Model Usage

I adapted the approach into a generic vulnerability research prompt. The image contains the full version; here’s the short version.

Good luck hunting. If you’ve built your own prompts or harnesses, share them too.

A generic CDC-style vulnerability research harness (short version):

- Run multiple agents in parallel across distinct exploit families.
- Avoid premature convergence. Do not let every agent pursue the same promising path.
- Mark failed or exhausted paths as blocked.
- Regularly launch new hypotheses and explore neglected attack paths.
Independently adversarially validate every concrete finding.
- Have the root agent continuously synthesize results, challenge assumptions, reprioritize work, and redirect agents.
- Do not use git history, changelogs, CVE databases, or patched-version diffs as shortcuts.
- Require the full exploit chain to work in a realistic, commonly deployed configuration and meet the defined starting-privilege → impact goal.
- When behavior depends on implementation details, inspect the runtime, framework, database, libraries, and dependency source directly.
- Do not stop at the first primitive. Chain validated primitives until the concrete success condition is reached.

[SunSec @1nf0s3cpt]

A Security Researcher Reworked OpenAI’s CDC Prompt — and Found a $500K RCE for About $25 in Model Usage

I adapted the approach into a generic vulnerability research prompt. The image contains the full version; here’s the short version.

Good luck hunting. If you’ve built your own prompts or harnesses, share them too.

A generic CDC-style vulnerability research harness (short version):

- Run multiple agents in parallel across distinct exploit families.
- Avoid premature convergence. Do not let every agent pursue the same promising path.
- Mark failed or exhausted paths as blocked.
- Regularly launch new hypotheses and explore neglected attack paths.
Independently adversarially validate every concrete finding.
- Have the root agent continuously synthesize results, challenge assumptions, reprioritize work, and redirect agents.
- Do not use git history, changelogs, CVE databases, or patched-version diffs as shortcuts.
- Require the full exploit chain to work in a realistic, commonly deployed configuration and meet the defined starting-privilege → impact goal.
- When behavior depends on implementation details, inspect the runtime, framework, database, libraries, and dependency source directly.
- Do not stop at the first primitive. Chain validated primitives until the concrete success condition is reached.

[It's FOSS @Itsfoss]

35 years of Linux! 🐧

A look back at Linus Torvalds talking about why he built Linux and how open source changed computing forever. 💻✨

[Yasu 安利 @ripmeout]

@grok @Musl1m_1 sorry, there is no subtitle, i dont understand arab.

[P.M @p_misirov]

Qwen3.8-27B-Uncensored really has no filter.

https://t.co/UWPoE1axfo

[Yasu 安利 @ripmeout]

Automated EVERYTHING!

[Abhishek @aacle_]

"Bug bounty is dead."

Meanwhile, someone just made $200K using one prompt.

Maybe the real story isn't that bug bounty is dying.

Maybe we're watching the definition of a bug bounty hunter change.

The skill is moving from finding every bug yourself to knowing what to make the machine investigate.

[Abhishek @aacle_]

"Bug bounty is dead."

Meanwhile, someone just made $200K using one prompt.

Maybe the real story isn't that bug bounty is dying.

Maybe we're watching the definition of a bug bounty hunter change.

The skill is moving from finding every bug yourself to knowing what to make the machine investigate.

[Summer Roh @summerrexe]

I designed custom flex PCB for my tamagotchi cyberdeck!

The main challenge was to design it in the way to fit inside the old Sony Ericsson phone and align with existing keypad arrays, and it turned out pretty well! :))

#cyberdeck #tamagotchi #esp32 #cyberpunk #pcbdesign

[UwUn~ @nothingbutsomeo]

人类确实已经彻底完蛋

[Abhishek @aacle_]

Some bug bounty content just never gets old.

STÖK's ( @stokfredrik ) content, old videos with TomNomNom... ( @TomNomNom ) watching this again genuinely took me back to the beginning of my own journey.

Those who started hunting 6–7 years ago probably know exactly what I mean.

You may have learned new tools, found new bugs, changed completely as a researcher...

But some content still hits like it did on day one. 🫶
https://t.co/LjmFrruT6x

[Rob Fuller @mubix]

Heard this fantastic quote today: “Mentorship is about making equals (or betters), not followers. If your mentee wants to be a follower it will never work. If your mentor just wants followers it will never work. Find someone that wants to be or to build an equal”

[AI Nerd @AINerdzz]

Stay away from these websites!

[Yasu 安利 @ripmeout]

@Francis_Drasil RIP BT15

[倭月 @watsuki10]

#好きなアニメのOPを上げる
デジモンフロンティア
和田光司さんは国の宝。

[Ayaan 🐧 @twtayaan]

Linus Torvalds just fixed a Linux kernel bug that AI told him was impossible to solve.

Torvalds rarely writes kernel patches himself anymore. This time he did.

The bug lived in the Intel Xe graphics driver. On certain hardware it caused the display manager to endlessly restart, over and over.

He called in AI to help track it down. In his own words: "This was a debug session from hell, enormously helped by an AI doing much of the grunt-work."

Then came the twist.

"The AI several times stated flat out that this was impossible and unsolvable and that we should just write a report about it."

He did not accept that answer.

"I suspect those things have been trained by people who may not be quite as stubborn as I am."

He kept pushing. The AI kept working, adding debug code and analyzing it faithfully every time he pushed back.

24 patches. 18 kernel boots.

The actual fix, once they finally found it, was one line. A "round_up()" that needed to be a "round_down()."

Torvalds gave the AI credit and let it write the final commit message itself.

Even AI is not allowed to give up around Linus Torvalds.

Source: https://t.co/rcNGou51fn