Twenty-four chapters of how I actually use Claude every day, across Claude Code and Cowork.
What's specifically in it:
When Opus 4.7 wins versus Sonnet 4.6 versus Haiku 4.5 in May 2026, including the SWE-bench gap (Sonnet 4.6 at 79.6, Opus 4.6 at 80.8) that finally makes Sonnet the default (Chapter 13).
A code-reviewer agent with confidence scoring and read-only tools, plus the GitHub Actions workflow that posts its output as a PR comment via gh pr comment(Chapters 7 and 16).
Token-cap recommendations per CI workflow type (20k for PR review, 40k for terraform plan review, 15k for build-failure triage) so a runaway 3am workflow doesn't surface as a surprise invoice (Chapter 16).
Two anti-pattern chapters from real failures. "Three failed fix attempts means the architecture is wrong, not the patch" is one of them (Chapters 23 and 24).
A daily documentation review wired to a Cowork scheduled task that runs /revise-claude-md against your engagement folders at 8:30am and posts a drift digest (Chapter 20).
Connector auth recipes for GitHub, Google Workspace, Microsoft 365, Slack, Notion, Zoom, Figma, Box, and Vertex AI on private GCP (Chapters 21 and 22).
The nine-step CLAUDE.md upgrade checklist (no command chaining, no AI-slop language, em-dash ban, decision-making section, debugging discipline, 2026 model references) that takes about 15 minutes and pays back in every subsequent session (Foreword and Chapter 2).
What it is not: a docs replacement. The Anthropic docs win on freshness. This is the practitioner companion that sits beside them.
Inside the book
Table of contents
How to read
How to Read This Book
Introduction
Before and After Claude
Foreword
Migration Guide (February to May 2026)
1
Setup & Configuration
2
Project Instructions (CLAUDE.md)
3
Plugins
4
MCP Server Integrations
5
Hooks & Automation
6
Slash Commands
7
Custom Agents
8
Custom Skills
9
Common Workflows
10
Prompting Guide
11
Agent Teams & Parallel Work
12
Self-Learning & Self-Optimisation
13
Claude Model Selection in 2026
14
Prompt Caching
15
Claude Code Desktop
16
Claude Code in CI
17
Claude in Microsoft 365 and Chrome
18
Claude Cowork: A Primer
19
Cowork Workflows for Client Reporting
20
Live Artifacts & Scheduled Tasks
21
Connector Authentication Reference
22
Vertex AI and Enterprise Deployment
23
Anti-Patterns (Claude Code)
24
Anti-Patterns (Cowork)
Closing
Closing note
A few pieces
From the pages
“
Be specific. "Use TypeScript strict mode" beats "write good code." "Mock at the service boundary, not at client internals" beats "test things properly."
Chapter 2: Project Instructions (CLAUDE.md)
“
Make the plan extremely concise. Sacrifice grammar for the sake of concision. At the end of each plan, give the user a list of unresolved questions to answer, if any.
These two rules cut planning overhead massively. Plans become checklists, not essays.
Chapter 2: Project Instructions (CLAUDE.md)
“
If the system cannot operate correctly, it must refuse to operate at all.
Chapter 23: Anti-Patterns (Claude Code)
“
If your first fix attempt doesn't work, STOP. Do not attempt a second fix. Go back to Phase 1: gather evidence.
Three failed fix attempts means the architecture is wrong, not the patch.
Chapter 23: Anti-Patterns (Claude Code)
“
Never claim a fix works until you have verified it end-to-end in the running system. Never deploy a fix and tell the user to check it. You verify it first.