Nixmac: Nix-darwin that speaks plain English

Taking Control of Toolbar Items in SwiftUI

SwiftUI bổ sung các API Toolbar mới giúp kiểm soát rõ ràng hơn về hiển thị và hành vi của toolbar items trên nhiều nền tảng, bao gồm visibilityPriority, ToolbarOverflowMenu, topBarPinnedTrailing, và toolbarMinimizeBehavior. Các tính năng này cân bằng giữa khả năng thích ứng của SwiftUI với sự điều khiển trực tiếp từ nhà phát triển trên iOS, iPadOS và macOS.

Nếu bạn đang phát triển ứng dụng iOS/iPadOS/macOS với SwiftUI và muốn tối ưu hóa giao diện người dùng bằng cách điều khiển chính xác các mục toolbar theo từng trường hợp sử dụng cụ thể, hãy đọc bài này để khám phá các API mới nhất giúp bạn kiểm soát tính năng hiển thị và tương tác của toolbar một cách linh hoạt.

Tamamon: A desktop pet that grows as you code with Claude Code

Tamamon is a macOS desktop pet app that grows and evolves based on your Claude Code activity. It features 20 collectible species unlocked via weekly gacha, interactive behaviors like feeding and playing, and real-time/weather reactions. Everything runs locally with no account, sign-in, or data uploads required.

The Dune keypad device can be your meeting controller and more

Project Mirage's Dune is a $149 three-key aluminum USB-C keypad for MacBooks that changes its button functions based on the active app — muting mics in meeting apps, triggering shortcuts in Excel or Chrome, and executing developer actions in VS Code or GitHub. It ships with a companion app for per-app shortcut configuration, calendar sync for meeting management, and Python scripting for custom automation. A Claude Desktop integration lets non-coders describe desired shortcuts in plain language and have Claude write and assign them automatically. A community marketplace for sharing 'skills' exists but is currently sparse. The device has no battery, draws power from the MacBook, and is built to fit flush against specific Mac models. Key drawbacks include overly sensitive buttons and a limited skill marketplace with no preview mode.

A new Linux driver gets us one step closer to daily driving the OS on an M3 MacBook

A new 10-patch series submitted to the Linux kernel adds support for built-in keyboards on Apple Silicon M2 and M3 MacBooks. The patches introduce an apple-dockchannel mailbox driver, a DockChannel HID transport driver, and tweaks to apple-rtkit and hid-apple drivers, enabling keyboard input when booting Linux on these machines. While Linux 7.2 already made it possible to boot on M3 MacBooks, input devices were previously unusable. The trackpad remains unsupported but is reportedly next in line for development.

I used GPT-5.5 to figure out why my laptop battery was draining, and it found the bug in minutes

After upgrading to macOS 27 Golden Gate Beta 2, an M4 Pro MacBook Pro experienced severe battery drain and overheating. Using GPT-5.5 via a coding agent harness called Pi, the author ran a chain of diagnostics that identified appstoreagent consuming 54-60% CPU due to an infinite retry loop in Apple Arcade metrics reporting. The root cause was a SQLite schema mismatch — a missing column (active_launch_events.recent_launch_times) caused repeated errors that triggered endless rescheduling. The fix was disabling the launchd agent with launchctl, which persists across reboots. The post highlights how AI coding agents with terminal access can be repurposed for system troubleshooting, not just software development.

Gaslight macOS Malware Is a Warning Shot at the AI Security Stack

The Gaslight macOS malware, attributed to a North Korean-linked threat cluster, is a Rust-based backdoor notable for embedding 38 fabricated prompt injection messages designed to confuse LLM-assisted malware triage tools. While it hasn't successfully bypassed any production AI analysis platform yet, it represents a deliberate iteration on earlier single-injection prototypes — signaling that adversaries are actively developing techniques to exploit AI-assisted security workflows. The attack targets the output stage of analysis pipelines rather than sandboxes, pushing LLM agents to abort sessions before flagging malicious behavior. Defenders are advised to treat malware content as adversarial input, sanitize raw strings before they reach model context, and maintain human review at the output stage. Specific IOCs including a LaunchAgent label, Telegram C2 traffic patterns, and SHA-256 hashes are provided.

Malware authors subvert AI detection systems

A new macOS backdoor called macOS.Gaslight, attributed to North Korean threat actors, embeds prompt injection code designed to make LLM-assisted security analysis tools abort or refuse to analyze the malware. SentinelLabs documented this technique, noting it is not the first such case — Check Point reported a similar approach a year ago, and Socket found comparable evasion in another payload. As AI-assisted malware analysis becomes standard practice, defenders should expect more malware built specifically to exploit and subvert these AI-based detection systems.