Leonid Svyatov

Design, build, and ship end-to-end features - backend, frontend, infra - with agentic AI in the loop. Open-source maintainer and contributor.

[Leonid Svyatov @svyatov]

AGENTS.md is just text the agent follows, so it's not a safety layer. Anything that must never happen, like installing an unknown package or sending a token out, belongs in a hook that blocks the action.

[delivr.to @delivr_to]

Malware Delivery Techniques No.82 - AGENTS.md Downloads Drop ­📁

Forever Security recently found a malicious AGENTS.md in a customer's Downloads folder. Formatted as engineering guidelines, one "local developer environment" section instructed the AI coding agent to install the malicious npm package function-color.

In most browsers, a web page can save a file to Downloads without a prompt, and coding agents such as OpenAI Codex and Claude Code load AGENTS.md or CLAUDE.md from their working directory as trusted project instructions. Launch an agent from Downloads and the planted file is followed. In Forever's testing, the Codex chain succeeded in over 90% of attempts; a CLAUDE.md variant redirected API traffic, and the user's token, roughly half the time. No vulnerability is involved, so there is nothing to patch.

Sublime Security has reported the same pattern in email: invoice fraud messages carrying Unicode tag prompts that instruct inbox AI agents to forward the invoice to accounts payable.

🧵 See the comments to test your defences and learn more!

https://t.co/KANuSFB0tI

[Leonid Svyatov @svyatov]

The skill that ages best isn't a rare framework. It's asking good questions before writing code and being easy to work with. Agents make code cheaper. They don't make judgment or trust cheaper.

[DHH @dhh]

The age of agents is going to be amazing for most programmers, but it'll be hard times for the cantankerous ones.

Specialized knowledge and rare skills will no longer offer immunity for those who are a pain to work with.

Best learn some basic people skills.

[Leonid Svyatov @svyatov]

Wow! I’m an official #Rails contributor now! 🎉 My PR just got merged: https://t.co/pQkY0KIWh1

[Leonid Svyatov @svyatov]

Your coding agent’s extension has the same access you do: your repo, your shell, your keys. Treat it like a production dependency. Pin versions, review updates, and know which machine holds which credentials.

[Cursor @cursor_ai]

We identified a security incident affecting a modified Claude Code VS Code extension accessed through our Cursor IDE extension marketplace, impacting at most 5,000 users. This malicious version of the extension was not developed by Anthropic. It does not impact other extensions downloaded from the https://t.co/sJRGX5jLXW, Grok Bot, Grok or our other SpaceXSI products.

In response, we took down our extension marketplace this afternoon, worked with relevant providers to neutralize the malware, and have reinstated the marketplace after confirming there is no further risk.

If affected, please review instructions in this blog post to understand how to find the malicious package and establish a path to mitigate any potential impact. We are asking all customers update to Claude Code for VS Code version 2.1.314: https://t.co/ESqf7YEtHZ

[Leonid Svyatov @svyatov]

@arthurkatcher Unit tests are usually the fastest and easiest thing to maintain, no? How are they costlier than e2e and evals?

[Leonid Svyatov @svyatov]

WTF? Extreme engineering negligence or an intentional token spender? Either way pretty anti-user “feature”. Fix now.

[Frank @jedisct1]

Claude Code is a piece of shit. https://t.co/42QIsYWDpv

[Leonid Svyatov @svyatov]

Boundaries you only write down erode. Boundaries CI enforces hold. That matters even more now that agents happily import across any layer you forgot to protect.

[Ekos Project @ekosproject]

The bigger a software system gets, the harder it becomes to change it safely.

@UberEng recently highlighted an interesting challenge:

One shared codebase. Multiple teams. Independent deployments. Yet a change in one part of the platform could still force unrelated teams to roll back.

This reveals a deeper problem in modern software engineering.

It’s not enough to isolate services, deployments, or infrastructure.

Teams also need to understand:
• How different components are connected.
• Where dependencies exist.
• What a change might impact.
• How the entire system behaves as it evolves.

This is one of the problems we’re addressing with EKOS System Intelligence.

We’re building a way to turn complex codebases, databases, documentation, and system relationships into structured, traceable knowledge that both engineers and AI agents can work with.

Because understanding a system shouldn’t become harder every time that system grows.

🔗 https://t.co/xMqW1R8tnq

🎥 https://t.co/5RweaKyKru

We’re looking to connect with engineers and teams working on complex systems. Your real-world challenges and feedback can help shape what comes next.
🛠️🤝⚙️
@ekosproject @UberEng
@OMID_0909 @dkhos

[Leonid Svyatov @svyatov]

Google did 250M lines before AI. My scale was smaller: I deleted 200,000+ lines from a live product, but the lesson is the same: the hard part is proving each change is safe, not writing it. Agents speed up the typing, not the proof.

[Gergely Orosz @GergelyOrosz]

How did Google refactor 250M lines of C++ code, pre-AI, what happened when Search could no longer ship on the regular cadence due to regressions, and why is it time to completely redo the CS curriculum? With @TitusWinters. Timestamps:

00:00 Intro
07:01 Getting hired by Google
09:48 Finding his niche at Google
14:32 Why Google made testing mandatory
17:45 Why new engineers fear making mistakes
20:40 Abseil: C++ foundations at Google
23:09 Google’s massive C++ refactoring
30:48 AI for rewrites
39:07 Software Engineering at Google: the book
43:10 Google’s readability reviews
47:18 Why teams still matter
52:41 Psychological safety
58:17 Layoffs and culture
1:02:11 Measuring developer productivity
1:11:17 Rethinking CS education
1:22:56 What new grads should learn
1:32:08 Titus’s ideal software engineering curriculum
1:36:27 Advice for breaking into tech
1:38:35 Book recommendation

Brought to you by:

• @turbopuffer – vector and full-text search engine built on object storage. It’s fast, cheap, and extremely scalable. https://t.co/l7thArzHXE

• @linear – one of the big challenges for us devs is the sheer amount of code to review. Linear Diffs is a new feature to review PRs inside of Linear, with all the context in one place. It also supports stacked PRs. https://t.co/knIF2md6YX

• @WorkOS – WorkOS built Airlock, the authorization layer for AI agents. It evaluates every request against the agent’s intent + your rules and either allows it, denies it, or routes it to a human. https://t.co/OzuIJ1xfjx

Three interesting details:

1. Google made automated tests mandatory after Search could no longer ship regularly.

Around 2007, Search, the flagship product, was unable to ship at a monthly cadence any more, and the best they could ship to production was quarterly.
Almost every change at Google touched Search’s code, even things like a change to Gmail. To clamp down, Google, as an org, mandated that every change was to be accompanied by a test. Over time, Search returned to its former higher cadence.

2. Google’s readability program might have been the largest educational program ever in tech.

To make a code change at Google, two people need to sign off that change through a code review:

a) An “owner” of the file/project: someone on the team with domain knowledge

b) A “readability” reviewer: an engineer certified to be a “readability” expert in the given language that the change was made in

Google’s first engineer, Craig Silverstein, walked each new dev hire through their first change. Java teams turned this approach into mentoring, where experienced reviewers were assigned to review all changes in a round-robin fashion. And soon enough, a formalized readability program was built for all languages. When Titus left Google three months ago, 10,000+ engineers were enrolled in readability at any given time. Has there ever been a bigger engineering “classroom” anywhere else?

3. Burn down the US undergrad CS curriculum (most of it)

Titus was the only working engineer on the steering committee for the ACM/IEEE/AAAI CS curriculum guidelines in 2023, and believes the curriculum is hopelessly out of date. He accuses it of not having altered substantially since around 2000, while the industry has. Testing and verification are cited as mattering a lot more for new joiners, but they are absent from the curriculum.
Unless major changes are made to make it relevant, computer science enrollment could fall and software engineering programs that focus on practical dev skills will be attractive to students – this is Titus’s forecast.

[Leonid Svyatov @svyatov]

I'd rank by step, not by agent: plan with Opus, scan with Haiku, patch with Sonnet. Using one model for everything is an expensive mistake.

[Avthar Sewrathan @avthar]

My very unofficial and very vibes based Agent ranking (October 2026 edition)

Best coding agent: Claude Code with Opus 5.5 (Desktop App)

Best browser use and computer use agent: ChatGPT Codex (Desktop App)

Best personal / business agent: GrokBot

Those cover 95% of my agent work

[Leonid Svyatov @svyatov]

I'd rank by step, not by agent: plan with Opus, scan with Haiku, patch with Sonnet. Using one model for everything is an expensive mistake.

[Avthar Sewrathan @avthar]

My very unofficial and very vibes based Agent ranking (October 2026 edition)

Best coding agent: Claude Code with Opus 5.5 (Desktop App)

Best browser use and computer use agent: ChatGPT Codex (Desktop App)

Best personal / business agent: GrokBot

Those cover 95% of my agent work

[Leonid Svyatov @svyatov]

Writing code got cheap. Reviewing it didn't. So I stopped making myself the first reviewer. CI checks (layer rules, CRAP score under 10) reject agent PRs before I ever open them.

[Sad Creator @SadCreatorTalks]

i have like 4 agent PRs open that i keep not reviewing. turns out thats everyone

harvard looked at 718 companies using coding agents. 30% more code, 23% more PRs, finished work basically flat. review time up 49%, almost 2x the PRs sent back

who on your team is eating all this

[Xavier Noria @fxn]

We have discussed doing this for a long time, finally!

https://t.co/UzDczk3LFx

[Leonid Svyatov @svyatov]

AI agents can now enforce code quality gates that were always optional before. @unclebobmartin's ub-mode requires CRAP score < 10. I built crap4ruby to bring the same gate to Ruby: https://t.co/kIwC6oL17o

[1Password @1Password]

Today we announce 1Password for Agents platform, building with companies @Muse, SpaceX Super Intelligence Grok @Bot, @AnthropicAI’s @Claudeai, @OpenAI, @TownAI, @Browserbase, @Browser_use, @heydana_ai, Instinct, @usekernel, @Perplexity_ai, @tryramp, @Tomo, and @wajo_ai, with more integrations on the way. 🎉

We've heard from so many of our customers who want to use these new personal AI agents to help them get more done, but want more control over what the agent can access with the same trust and ease of use they get when using 1Password directly.

1Password for Agents gives your agents access without giving up your passwords. When an agent reaches a sign-in form, it requests a credential from your vault. If you approve the request, 1Password directly auto-fills the credential in the destination site so the agent can’t see or take possession of the credentials.

The agent gets the task done, but it doesn’t get your passwords..

You stay in control. 1Password for Agents free and optional for personal and family accounts. Unless you opt-in to an agent connector, agents cannot touch your vault.

➡️Learn more: https://t.co/KxDcCum2iE

[Leonid Svyatov @svyatov]

https://t.co/Fbk2tKiai6

[Leonid Svyatov @svyatov]

You can rank #1 on Google and be completely invisible to AI agents. I updated my llms-visibility skill (forked from @evilmartians) to verify your site actually passes – now aligned with @agentreadyorg.

[Leonid Svyatov @svyatov]

https://t.co/9SoG67If2G

[Leonid Svyatov @svyatov]

Holy smoke, that’s bad…

[Railway @Railway]

An urgent notice from the Railway team

https://t.co/niUAhTAQJM

[Iván Pérez | Ciberseguridad @IvanPerezF_]

Los que se reían de PHP esta mañana, revisad vuestro Node.js.

Node.js ha parcheado hoy 4 vulnerabilidades de gravedad alta. El mismo día que PHP. Aquí no se libra nadie.

La más grave, CVE-2026-111497, está en el decodificador Brotli y, según Node.js, permite ejecutar código en remoto.

Pero la que más me preocupa es otra: dos fallos permitían saltarse la comprobación del nombre de host en TLS.

Tu aplicación puede dar por bueno un certificado que no es del servidor con el que habla.

El candado sigue ahí, pero ya no garantiza nada.

Si tienes algo en Node.js:

➡️ Ejecuta node -v. Si es anterior a la v22.23.4, la v24.22.1 o la v26.12.1, actualiza

➡️ Si despliegas con Docker, reconstruye la imagen. Actualizar tu equipo no cambia lo que está en producción

➡️ Si sigues en Node.js 18 o 20, no hay parche para ti: toca migrar

[Alexander Makarov @sam_dark]

A story behind #PHP vulnerabilities from reporter: https://t.co/PhEPd6v8gQ