Nikita Ermolenko

host of @readaggregator telegram channel

[Nikita Ermolenko @otbivnoe]

There're so many interesting stuff, so I don't just save it to bookmarks, I do share it in my tchannel! https://t.co/4DKJyKAdVz #Swift #iOS

[Cheng-Yuan (Sam) Lee @cl571128]

We've received several questions about the Opus 5 FrontierCode results, where scores decline as reasoning effort increases. In fact, the behavior is expected under the benchmark design. FrontierCode evaluates merge-ability rather than correctness alone, incorporating criteria that reflect user experience.

One such verifier is a scope criterion, which penalizes modifications to the codebase beyond what the task requires. We observe this effect across all frontier models we evaluated, though it is most pronounced in Opus 5: at higher reasoning efforts, the model shows a stronger tendency to refactor code unprompted.

[Kenton Varda @KentonVarda]

Building software by prompting agents without ever reading or editing the code directly isn't actually a new thing. It's just called being an engineering *manager*. We have decades of experience about how to be effective as engineering managers. It is a different skill, though.

[Codex Releases @CodexReleases]

Codex CLI 0.145.0 is out!

Highlights:
- Multi-agent V2 is now stable, with configurable sub-agent models, reasoning levels, and concurrency.
- Experimental paginated thread history adds efficient resume, occurrence search, persisted names, and memories.
- Audio inputs and tool outputs land, with streaming realtime V3 conversations.
- / import expanded to migrate Cursor and Claude Code settings, MCP servers, plugins, sessions, and memories.

Complete details in thread ↓

[Mysk 🇨🇦🇩🇪 @mysk_co]

Introducing Loupe, our latest privacy app for iOS. Discover what apps can learn about you just by reading data your iPhone already exposes, such as your languages, installed apps, device sensors, and much much more

Loupe is free, private, and open source. Give it a try 👇

[Matthew Berman @MatthewBerman]

94% chance the Codex quota will be reset within 48 hours.

https://t.co/oJDOzAMNSX

[Brandon Butch @BrandonButch]

iOS 27 adds separate volume sliders for Ringtone, Alarms, and Alerts!

[Artem Novichkov @iosartem]

Learn how to name your tasks in Swift Concurrency. The names show up in LLDB, Instruments lanes, and SwiftUI's .task modifier. One small string, much better debugging.

https://t.co/bTmpwF4Rtm

[Axel Le Pennec @alpennec]

Apple updated its Human Interface Guidelines for Sheets on March 24, 2026 to add guidance for button placement in the toolbar.

"In iOS & iPadOS, the Cancel button belongs on the leading edge of the top toolbar."

Time to update my apps 🙂

Documentation: https://t.co/y2pRhemXQJ

[Nikita Ermolenko @otbivnoe]

@zach_bai @warpdotdev it would be awesome to also have horizontal tabs within each vertical tab line cmux supports 🙏

[Nikita Ermolenko @otbivnoe]

@warpdotdev Could you also support horizontal tabs inside each vertical tab as in cmux? For each worktree I want to have a bunch of tabs: lazygit, codex… Yea, I can now have it using cmd+D, but this way they’re not visible

[am.will @LLMJunky]

🚨 Critical Supply Chain Attack

LiteLLM (a core dependency for 100s of AI tools) was compromised in PyPI versions 1.82.7 and 1.82.8.

The malicious litellm_init.pth file runs on ANY Python startup, steals SSH keys, env vars, cloud creds, Kubernetes secrets, etc., then exfiltrates them.

It even tries to persist via systemd/K8s.

If you use LiteLLM, check immediately:

- Run: pip show litellm
- If version 1.82.7 or 1.82.8 -> uninstall + rotate ALL credentials NOW
- Pin to safe version: litellm<=1.82.6 (or latest clean release)

Also purge caches: pip cache purge

[Mike Rundle @flyosity]

--dangerously-skip-permissions

[Cline @cline]

Bring your ChatGPT subscription to Cline for inference.

We partnered with @OpenAI to let you use your existing subscription. Sign in and access all the models in your subscription. No API keys, flat-rate pricing instead of per-token costs.

Here is how to enable this:

[Nikita Ermolenko @otbivnoe]

@7andrees @Dimillian Have you found a solution? probably faced the same problem

[Peter Steinberger 🦞 @steipete]

Some good tips in there, also A LOT of workarounds to fight Sonnet's post-training choice for rushing instead of being diligent and careful. With codex you can drop 90% of the charade. https://t.co/IuRx9AX8wR

[Swift Language @SwiftLang]

Swift 6.2 is here, including so much 🆕🥳:
⏩Approachable concurrency
🦺Opt-in strict memory safety
📦WebAssembly support
🏃Faster build perf for macros
🐞Enhanced async debugging
📚Inline stack-based arrays
🚇Subprocess support
🧪Test result attachments
https://t.co/sQrluRKcJQ

[Nikita Ermolenko @otbivnoe]

@iosartem you play. llm code

[Nikita Ermolenko @otbivnoe]

@iosartem The hunching dog days are over

[Mattt @mattt]

Apple's JS-rendered docs leave AI tools staring at blank pages. I got tired of Claude writing broken SwiftUI code, so I built https://t.co/pb1UMcV8mC

Swap developer․apple․com → sosumi․ai in URLs to get clean, AI-readable Markdown. Or connect via MCP.

[Claude @claudeai]

New in Xcode 26 Beta 7: Users can now log into their Claude account from Xcode.

Generate code and documentation, automatically fix errors, and create inline playgrounds using coding intelligence features in Xcode, powered by Claude Sonnet 4.