Claude Code Up and Running - Harness the Power of Agentic Coding (Early Release) (Ken Kousen)（Z-Library）

【One-Line Pitch】 A terminal-first field guide to Anthropic's Claude Code that teaches the durable practice of directing an autonomous coding agent—not just its flags. Best for developers, tech leads, and architects already comfortable in the shell who want to understand what changes when an agent can work inside a real project. 【Book Arc】 - **Opening (~0%–9%)**: Frames the shift from autocomplete to agentic coding and introduces Agent-Driven Development (AgDD)—the agent drives production work while you drive direction, verification, and judgment. Solves the "what actually changes?" question before any tooling. - **Early (~9%–28%)**: Sets expectations for who should read and how to use the book: cumulative chapters, a companion examples repo organized by app and chapter snapshot, and "Ask Claude Code" sidebars modeling pull-mode learning. - **Early–Middle (~28%–44%)**: Chapter 1's core argument—agentic tools blur language, framework, and role boundaries, so the bottleneck becomes direction and verification. Introduces the doghouse/cabin/skyscraper scale of software complexity and shows how methodology must change with stakes. - **Middle (~44%–53%)**: Defines Claude Code as a harness around a model (loop, terminal interface, project context, tool calls, permission gates, MCP, hooks, skills) and distinguishes it from chat tools and IDE autocomplete. Surveys where it runs—terminal, desktop, web, IDE, GitHub, mobile—while insisting the terminal remains canonical. - **Late (excerpts do not cover)**: Table of contents points to permissions and trust boundaries, skills and hooks, subagents and agent teams, working in inherited codebases, MCP integrations, prompt-to-PR workflows, and review/testing/debugging/failure recovery. - **Ending (excerpts do not cover)**: A closing chapter on surfaces beyond the terminal, model/provider choices, and cost/context trade-offs—what stays stable as tools keep moving. 【Key Takeaways】 - **Agent-Driven Development (AgDD) is the book's spine** (Opening): the agent produces, you direct, verify, and judge. The more capable the agent, the more your judgment matters—not less. - **The harness, not the model, is where the patterns live** (Middle): the agent loop, context assembly, tool calls, permission gates, MCP, hooks, and skills are the durable layer; the model is a knob. - **Complexity has three scales—doghouse, cabin, skyscraper** (Middle): small weekend projects tolerate vibe coding; anything someone depends on demands trust contracts, boundaries, and a testable core; enterprise work adds coordination, CI/CD discipline, and reusable workflows. - **Discipline is set by stakes, not size** (Middle): purpose, exposure, longevity, and accountability determine how much process applies—a modest service can still require cabin-level rigor. - **Claude Code is agentic from the start** (Middle): unlike line-by-line autocomplete, it plans multi-step work, decides file order, and checks back at permission points. Conversation is a side effect of the work. - **The surface is the interface, not the product** (Middle): terminal, desktop, web, IDE, GitHub, and mobile are entry points; sessions can sometimes hand off between them, but the underlying loop stays the same. - **Pull-mode learning is the intended habit** (Early): ask "What is this?", "Why this way?", and "What are alternatives?" at the moment of need—efficient, but incomplete, so fill gaps deliberately. - **Teach the workflow, verify the details** (Early): commands, pricing, and model names drift; the book deliberately teaches durable practice and defers fast-moving specifics to current docs. 【Reading Tips】 - Read the first chapters in order—the book is intentionally cumulative, and decisions that seem fussy in the doghouse chapters become non-negotiable at skyscraper scale. - Type along with the tutorial chapters and use the companion repo (organized by app first, chapter snapshot second) rather than reading passively. - Treat "captured" prompts as records from the author's machine; run the "try this now" prompts yourself and expect different results. - Skim the surface survey (terminal vs. desktop vs. IDE vs. web) if you already know where you'll work; deep-read the AgDD framing and the complexity-scale discussion, which anchor everything later. - Use the "Ask Claude Code" sidebars as a template for your own habits, not as an oracle—verify explanations before trusting them. 【Coverage Limits】 This guide is based on the front matter and early chapters of an Early Release draft; later chapters (permissions, skills and hooks, subagents, MCP, prompt-to-PR, debugging) are listed in the table of contents but their content is not covered by the excerpts.