The ultimate field guide for mastering agentic coding with Claude.
From your first /help to orchestrating 10-agent pipelines.
Concepts · Best Practices · Tutorials · Techniques · Examples
This is a practical, opinionated knowledge base for engineers who want to use Claude Code not just as an autocomplete tool, but as a genuine development partner — one that can plan, implement, test, debug, and review entire features with minimal hand-holding.
The difference between a casual Claude Code user and a power user isn't access to secret features. It's knowing how Claude Code actually works — its context window, its tool execution model, its agent dispatch mechanism — and designing your workflows around that understanding.
This guide covers all of it.
New to Claude Code? Start with the companion workspace template: claude-code-onboarding — a ready-to-clone project with CLAUDE.md, rules, skills, hooks, and agents pre-wired so you can skip the setup and start using the patterns from this guide immediately.
The building blocks you must understand before anything else clicks.
| Concept | What It Is | Why It Matters |
|---|---|---|
| CLAUDE.md | Project instruction file auto-loaded into every session | Your single biggest lever for consistent behavior |
| Subagents | Specialized agents Claude can spawn to parallelize work | 10x throughput on multi-part tasks |
| Skills & Commands | Custom slash commands and reusable behavior packs | Encode expert workflows once, reuse forever |
| Hooks | Shell scripts that fire on Claude Code lifecycle events | Enforce quality gates without human intervention |
| MCP Servers | External tool servers that extend Claude's capabilities | Give Claude access to docs, DBs, browsers, APIs |
| Memory | Persistent cross-session knowledge store | Claude remembers your preferences and project context |
| Tools & Permissions | The tool execution model and permission modes | Understand what Claude can and can't do |
| Modular Rules | 3-layer rules architecture: CLAUDE.md → .claude/rules/ → lessons.md |
Keep instruction files manageable as projects grow |
| Agent Definitions | YAML frontmatter schema for custom .claude/agents/ files |
Define reusable specialized agents with scoped tools and behavior |
| Claude Code Desktop | GUI with embedded browser, live preview, and PR automation | Visual testing and session mobility on top of the same CLI engine |
Hard-won patterns from real agentic workflows.
| Guide | Summary |
|---|---|
| Context Management | Keep Claude sharp across long sessions and multi-step tasks |
| Prompt Engineering for Agents | How to write instructions that produce consistent, correct behavior |
| Workflow Design | Structure tasks so Claude can execute them autonomously |
| Security | Prevent prompt injection, credential leaks, and scope creep |
| CLAUDE.md Design | How to write a CLAUDE.md that actually changes behavior |
| Error Recovery | Build systems that fail gracefully and recover predictably |
Step-by-step guides from zero to shipped.
| Tutorial | Level | What You Build |
|---|---|---|
| Getting Started | Beginner | Your first productive Claude Code session |
| Your First CLAUDE.md | Beginner | A project instruction file that shapes every session |
| Building a Custom Skill | Intermediate | A reusable slash command for your workflow |
| Wiring Up Hooks | Intermediate | Automated quality gates on every session |
| Multi-File Refactor with Agents | Advanced | Orchestrating parallel agents on a large codebase |
Advanced patterns for agentic engineering.
| Technique | Description |
|---|---|
| Parallel Agent Orchestration | Launch independent agents simultaneously, merge results |
| Subagent-Driven Development (SDD) | The Spec → Implement → Review pipeline |
| Plan-Execute-Verify | Approval-gated implementation for high-stakes changes |
| Context Injection Patterns | Get the right information into Claude's window at the right time |
| Self-Improving Codebases | Lessons, rules, and memory loops that make Claude smarter over time |
| Worktree Isolation | Parallel branches without merge conflicts |
| Data-Driven Hooks | Move enforcement rules from scripts into markdown files — add rules without editing code |
| Autonomous Loops | Stop hook that feeds the prompt back — Claude iterates until a verifiable promise is fulfilled |
Ready-to-copy reference implementations.
| Example | Description |
|---|---|
| CLAUDE.md — Monorepo | Full CLAUDE.md for a multi-stack monorepo |
| CLAUDE.md — Single Service | Minimal CLAUDE.md for a single FastAPI service |
| Pre-commit Hook | Block commits that violate code standards |
| Stop Hook | Auto-append session log on exit |
| Custom Skill | A /deploy skill that runs the full deploy pipeline |
| Rules File Set | A complete .claude/rules/ directory |
Most people treat Claude Code like a faster Stack Overflow. Power users treat it like a junior engineer who needs good onboarding.
The key shifts:
1. Invest in CLAUDE.md. Every hour you spend writing clear project instructions saves 10 hours of corrective prompting. Claude reads it at the start of every session — make it count.
2. Design for autonomy, not assistance. Write tasks as success criteria, not step-by-step instructions. "Make all 47 tests pass" beats "first open auth.py, then find the token function, then...".
3. Use subagents for parallelism. If a task has 3 independent parts, dispatch 3 agents. Claude Code's agent system is the closest thing to having a junior dev team.
4. Build your rules library. Every time Claude makes the same mistake twice, that's a rules file entry. At the third occurrence, it becomes a permanent rule. Your .claude/rules/ directory is institutional memory.
5. Let hooks enforce standards. Quality gates enforced by hooks are better than quality gates remembered by humans. Use the hook system to automate what you'd otherwise forget.
- Claude Code Documentation — Official Anthropic docs
- Claude Code on GitHub — Source, issues, changelog
- Anthropic Discord — Community discussion
- Anthropic Research — The science behind the model
- Claude Code Onboarding — Workspace template with pre-wired CLAUDE.md, rules, skills, hooks, and agents to bootstrap a new project
Found an outdated pattern? Have a technique that's saved you hours? PRs welcome.
- Keep entries practical — real code over theory
- Link to official docs where they exist
- Mark opinionated patterns clearly