Claude Code Design Skill

Source-grounded answers about how Claude Code and adjacent Anthropic agent systems are designed. Pulls from a NotebookLM RAG (Claude Code Design notebook) when available, and from curated local references otherwise.

When to Use

Trigger when user asks about:

• Claude Code architecture, orchestration, or sub-agents
• Agent skill anatomy (SKILL.md, references, eval)
• Progressive disclosure, 200-line SKILL cap, hooks, worktree isolation
• Memory-first / determinism-over-probability patterns
• Karpathy self-improvement loop for skills
• CLAUDE.md layering (directives / orchestration / execution)

Do NOT trigger for generic LLM design questions unrelated to Claude Code / Anthropic agent patterns.

Retrieval Order (Do Not Skip)

1. Check the notebooklm library for an active "Claude Code Design" notebook: