Claude Radar mark: a C-shaped ring and a signal point. Not a vendor logo. Claude Radar

English

Topics

Customization

Skills, hooks, subagents, MCP, plugins, and AGENTS.md — ways to extend Claude Code for work you repeat.

Graphite cursor, cream paper, and a terracotta sphere on a wooden desk.

Guide

What to put in CLAUDE.md

Write the commands, conventions, and checks you would otherwise re-explain every session.

Build

Claude Radar guide card: How do you add a Claude Code hook that notifies you, formats an edit, or blocks a command?

Guide

How do you add a Claude Code hook that notifies you, formats an edit, or blocks a command?

A Claude Code hook runs a shell command at a fixed moment: a desktop notification when Claude is waiting, Prettier after an Edit or Write, or a PreToolUse check before a tool call. Exit code 2 blocks events that can block, including PreToolUse. On PostToolUse it only shows stderr after the tool has already run, and PermissionRequest and Notification ignore it. Exit 0 with JSON is structured control for that event — allow, deny, ask, or block — and exit 0 alone leaves a PreToolUse call in the normal permission flow.

Build

A long cream paper ribbon folds into two branches and carries a few small tools across an otherwise empty page, illustrating a million-token window that must hold many rounds of work.

YouTube

Why MiniMax Says Agents Need a Million-Token Window

Olive Song tells Thomas Wolf that MiniMax M3 is a bet on three things at once: coding and agent skill, a million-token context window, and vision. Her reason for the long window is specific.

Claude Radar brief card: Anthropic opens a paid-plan portal for Claude plugins

X

Anthropic opens a paid-plan portal for Claude plugins

On 25 September 2026, Anthropic’s ClaudeDevs account and a product blog positioned plugins as the main third-party extension format for Claude and opened a directory submission portal to developers on paid Claude plans.

Overhead still life of a drifting compass among scattered papers, with a slim playbook entering from the side.

YouTube

Why MCP tools still left LinkedIn’s coding agents lost

Ajay Prakash, a LinkedIn software engineer, told an AI Engineer audience why wiring Model Context Protocol (MCP) search, docs, and tickets into coding agents still stalled on slightly complex internal jobs.