The Claude Power-User Handbook

Most Claude users write the same prompt twice. Power users write it once and ship a skill. This handbook is for developers who have Claude Code running and want to stop leaving capability on the table. It covers the full stack: how to write prompts that behave like software, how to structure context across system prompts, CLAUDE.md, and MEMORY.md, and how to apply the four-part prompt pattern that makes outputs reliable enough to package. From there, the book moves into Skills and subagents — the two primitives that make work compound. Readers learn the anatomy of a SKILL.md file, the test for when a prompt earns promotion to a Skill, and how to build a library that scales past fifty entries without becoming a maintenance burden. Subagent chapters cover agent file structure, isolation modes, and three composition patterns: parallel workers, sequential pipelines, and supervised dispatch. The book also covers what degrades results — prompt anti-patterns with a diagnostic checklist, the wrong tasks for extended thinking, and failure modes when composed flows get too deep. Every chapter is built around what actually transfers to a real workflow: naming conventions, versioning, lightweight evals, and a full skill lifecycle from first draft to retirement. One weekend of focused reading. Months of compounding output.