The Claude Code Playbook

playbooks

The Claude Code Playbook

The full playbook for shipping with Claude Code. Install, configure, optimize, deploy. Every step that turns a fresh terminal into a senior engineer.

Playbook / 30 Days / Claude Code

Claude Code: Zero to Operating System

30-day progression from first install to running sales, content, outreach, and engineering from one terminal. Each day builds on the last.

Architecture: Skills → MCP Servers → Hooks → Agents → Orchestration

Install Claude Code, run your first session, understand how it differs from claude.ai. Terminal-native AI with full codebase access and real file editing.

Content Angle

I gave Claude Code access to my entire codebase. Here's what happened.

“90 seconds. That's how long it took Claude to understand a 50k-line project.”

Key Concepts

• Terminal-native AI
• Full codebase access
• Real file editing (not suggestions)
• Git-aware out of the box

Build Examples

• First commit made entirely by Claude Code
• Codebase architecture review
• Automated bug fix from error log

Do This

Install Claude Code. Open a project. Ask it to explain the architecture. Watch it read every file. Then ask it to make a small fix and commit it.

Plan mode, extended thinking, ultrathink, piping, permission modes. Most people use 1% of what the CLI can do.

Content Angle

Most people use 1% of Claude Code's CLI. Here's the other 99%.

“Type 'ultrathink' in any Claude prompt. Watch what happens.”

Key Concepts

• Plan mode: think before executing
• Extended thinking / ultrathink
• Piping output into Claude
• Permission modes (yolo, ask, deny)

Build Examples

• Pre-commit AI linter
• Real-time log anomaly detector
• Build script with AI quality gate

Do This

Try plan mode on your next complex task. Use ultrathink for architecture decisions. Pipe your test output through Claude and ask it to summarize failures.

CLAUDE.md hierarchy, MEMORY.md auto-persistence, rules files. The 4-layer memory system that makes Claude Code remember everything.

Content Angle

Claude has memory. Here's how to actually use it.

“I tell Claude three things in CLAUDE.md and it never makes those mistakes again.”

4-Layer Memory System

• ~/.claude/CLAUDE.md (global identity)
• project/CLAUDE.md (project rules)
• .claude/rules/*.md (path-scoped)
• MEMORY.md (auto-persisted learnings)

Build Examples

• Personal AI identity system
• Client context auto-loader
• Monorepo path-scoped rules

Do This

Create your first CLAUDE.md. Tell it who you are, how you code, what to avoid. Watch consistency improve immediately. Then add one rules file scoped to your API directory.

@-imports, custom instructions per file type, context window management. The most important AI skill nobody teaches.

Content Angle

Context engineering is the most important AI skill nobody teaches.

“I import my client list, API spec, and style guide into every session automatically.”

Key Concepts

• @-imports: load live data into context
• Scope rules to file path globs
• Token budget management
• Dynamic context from external files

Build Examples

• Auto-loading API specs per directory
• Dynamic context from databases
• Smart context pruning strategies

Do This

Create 3 rules files scoped to different parts of your codebase. Watch Claude behave differently in each directory. Import one external reference file and see it used in context.

Build your own /commands, parameterized prompts, skill creator. Turn repetitive workflows into one-word triggers.

Content Angle

I type /proposal and Claude generates a full client proposal in 30 seconds.

“6 commands replaced 6 hours of weekly manual work.”

Key Concepts

• Reusable workflows as commands
• Argument passing with $ARGUMENTS
• Command chaining

Build Examples

• /content -- LinkedIn posts in your voice
• /proposal -- Client proposals with ROI
• /sop -- Standard operating procedures
• /client-prep -- Meeting preparation

Do This

Identify one thing you do every week. Turn it into a /command. Measure time saved. Then build two more. The compound effect is immediate.

Pre/post tool hooks, notification hooks, automated guardrails. The invisible layer that makes everything safe and observable.

Content Angle

Claude Code hooks: the feature that makes everything else 10x more powerful.

“I set up a hook that prevents Claude from ever committing secrets. Took 2 minutes.”

Key Concepts

• Run scripts before/after Claude acts
• Enforce rules automatically
• Pre-tool and post-tool hooks
• Notification on file changes

Build Examples

• Pre-commit security scan
• Auto-notification on file changes
• RTK token optimization hook
• Auto-format on save hook

Do This

Add one hook: a post-commit notification that tells you what Claude changed. Start small. Then add a pre-tool hook that blocks dangerous commands. Build guardrails, not gates.

Connect Claude to external APIs: CRM, LinkedIn, email, Slack. The Model Context Protocol gives Claude Code hands to touch the real world.

Content Angle

Claude Code just got hands. MCP servers let it touch the real world.

“Claude just sent a Slack message, updated my CRM, and drafted a LinkedIn post. In one command.”

Key Concepts

• Model Context Protocol (MCP)
• Tool bridging to external APIs
• API orchestration from terminal
• Read vs write action safety

Build Examples

• Attio CRM integration
• Unipile LinkedIn API
• Apollo lead enrichment
• Framer web publishing

Do This

Connect one MCP server. Start with Slack or your CRM. Watch Claude start taking real actions in the tools you already use. Then connect a second one and chain them together.

Full skill system: SKILL.md, references/, scripts/, multi-step workflows. Skills are reusable capabilities with their own knowledge base and tools.

Content Angle

Skills are Claude Code's superpower. Here's how to build one in 10 minutes.

“59 custom skills. Each one replaced a SaaS tool or a manual process.”

Skill Structure

• SKILL.md -- capability definition
• references/ -- knowledge base
• scripts/ -- executable tools
• Triggers and conditions

Build Examples

• carousel-maker (LinkedIn carousels)
• cold-dance (outreach sequences)
• post-to-pipeline (lead gen)
• nano-banana (image generation)

Do This

Create your first skill. Pick something you do manually at least once a week. Give it a SKILL.md, one reference file, and one script. Use it on a real task and iterate.

Subagent system, background agents, parallel execution, worktrees. Specialized agents for different tasks running concurrently.

Content Angle

I launched 4 agents in parallel. They finished in 3 minutes what would take me 2 hours.

“prospect-researcher, content-writer, security-auditor, fullstack-developer, all running at once.”

Key Concepts

• Specialized agent definitions
• Concurrent execution
• Background tasks
• Agent isolation with worktrees

Build Examples

• Prospect research + content gen in parallel
• Multi-agent code review
• Background monitoring agent
• Security audit while you build

Do This

Define 3 agent types for your workflow. Give each a clear scope and explicit boundaries. Run them on a real task. Watch the parallel execution.

Everything connected. The second brain as a business OS. Git-controlled knowledge base with Obsidian, PARA method, daily workflows, compound automation.

Content Angle

This is not a coding assistant anymore. This is how I run my entire business.

“59 skills. 36 commands. 12 agents. 9 MCP servers. One terminal.”

The Full Stack

• Second brain (Obsidian + Git)
• Claude Code (orchestration layer)
• MCP servers (external APIs)
• Skills (reusable capabilities)
• Agents (parallel execution)
• Hooks (automated guardrails)

Build Examples

• Daily command center (cockpit)
• Content-to-pipeline automation
• Full client onboarding in one command
• Automated prospect qualification

Do This

Map your entire workflow. Identify what's still manual. Build a skill for each manual step. Connect via MCP. Define agents for parallel work. Run it all from one terminal. You now have an operating system.

Claude Code vs Traditional AI Tools

Why Claude Code is a different category entirely.

Common Pitfalls

Avoid these to reach Level 9 faster.

No CLAUDE.md

Starting sessions with zero context. Claude repeats mistakes because you never told it your rules. Write it on Day 1.

Skipping hooks

Running agents without guardrails. One accidental delete or secret commit undoes hours of work. Set PreToolUse hooks early.

One-shot everything

Treating Claude like a chatbot instead of a system. Build commands and skills. Repetition should be one keystroke.

No agent boundaries

Agents without explicit 'you cannot' rules go rogue. Define what each agent can read, write, and call. Scope is safety.

Context window neglect

Cramming everything into one session. Use @-imports and rules files to load context on demand, not all at once.

No measurement

Hours saved, errors caught, leads enriched: if you're not measuring, you can't optimize. Log outcomes from Day 1.