Blog

Instead of one tool per message shape, @json-render/chat lets an LLM compose a Spec from a catalog and converts it into a Chat SDK card, posted to Slack as Block Kit. This article builds a Slack bot that posts tables and charts.

Chrome has implemented a `displaymode="icon-only"` attribute that renders `<camera>`, `<microphone>`, and `<geolocation>` as icon buttons without text. This article explains how to use `displaymode="icon-only"`.

Repeating a media query across CSS files means editing every copy when a breakpoint changes. @custom-media names a media query like --md so input and breakpoint presets live in one place. This article covers its syntax and specified behavior.

agent-desktop is a Computer Use CLI for AI agents from Vercel Labs. It reads the OS accessibility tree instead of screenshots and operates apps through refs. This article covers the basics, then has Claude Code test a SwiftUI app and fix a bug.

Reparsing the whole LLM response on every chunk recreates elements and loses state such as text selection. streamAppendHTML() returns a WritableStream that parses HTML onto the end of an element. This article covers its usage and sanitization.

Claude now supports MCP Tasks and Elicitation. Tasks turn slow tool calls into a task ID plus polling, and Elicitation asks the user for input mid-operation. This article builds an MCP server with the TypeScript SDK and calls it from Claude Code.

Safari 27.1 ships the Device Posture API and Viewport Segments API behind feature flags for foldables like iPhone Duo. This article explains how both APIs work, how they evolved, and how to build layouts that avoid the fold.

The OpenAI Decisions API returns condition checks, classifications, and graded ratings for an input. Using an example that evaluates support inquiries from TypeScript, this article explains when to use predicate, choice, and score.

MCP Events is an experimental extension that lets MCP servers notify subscribers of events via poll, push, or webhook delivery. This article gives an overview, then uses AI SDK's webhook support to subscribe, receive events, refresh, and unsubscribe.

Chrome DevTools MCP lets AI agents drive a browser, but the DOM can't tell them which component rendered an element or why it is slow. react-devtools-cdt-mcp registers React inspection and profiling tools with it. Let's try it with Claude Code.

Trusting AI-generated code requires understanding before implementation and verification after it. pstack is an agent skill collection that combines investigation, design, implementation, and review. This article covers its principles and usage.

Claude Code Mods are TypeScript/JavaScript functions that run inside the Claude Code process to observe, rewrite, or answer events. This article builds a mod that blocks dangerous Bash commands and explains how Mods work, their design, and security.