Anthropic's guide to steering Claude Code: CLAUDE.md, skills, hooks, rules, and subagents

驾驭 Claude Code：CLAUDE.md、技能、钩子、规则、子智能体等

Anthropic's official blog lays out five mechanisms for steering Claude Code: CLAUDE.md files as project-level instructions, skills for templated task execution, hooks that auto-trigger checks or scripts before/after actions, rules to constrain model behavior, and subagents that split complex work across independent workers. The post is a conceptual walkthrough with usage guidance—no benchmarks or pricing changes are disclosed.

Why it matters: Anthropic published a practical guide on steering Claude Code, breaking control mechanisms into five layers. It's a usage guide, not a product launch, so it doesn't hit 85. But it's substantive and precisely targeted at Claude Code users—worth featuring.