Chapter 1
The Agentic Developer Paradigm
This guide transitions you from using Claude Code as a glorified autocomplete to operating it as a true engineering multiplier. You will learn to manage context, structure iterative prompts, and use CLI commands to eliminate friction across your entire development lifecycle.
Efficiency Impact Analysis
Average weekly hours — manual dev vs. Claude Code-optimised.
Context is King
Over-stuffing context degrades quality. Point Claude to specific evidence, not the whole codebase.
Iterative Loops
Scope → Scaffold → Implement → Verify. One-shot complex tasks rarely survive first contact.
Tooling Synergy
Pipe test failures and lint errors straight back into the prompt. Build a self-healing dev loop.
Chapter 2
Command Mastery
Knowing when and why to use each slash command determines whether you're fighting the tool or flowing with it. Tap any card for the professional use case.
Chapter 3
Context Management
The primary reason AI coding assistants fail is context poisoning — irrelevant data degrading response quality. Managing token limits and relevance is an active discipline.
As a session grows, conversation history consumes your token budget. Eventually the AI loses track of early instructions and response quality degrades — without a clear signal.
- Monitor: Watch the token counter. Amber/red means you're at risk of context collapse.
- Compact:
/compactsummarises history while keeping file context and instructions intact. - Clear:
/clearfor a completely unrelated task — a blank slate beats polluted context every time.
Never ask Claude to "look at the whole project". Be the detective — point it to the exact evidence needed.
Weak: "Fix the routing issue."
Strong: "Review src/router.ts and src/views/Home.vue. Auth middleware
triggers a redirect loop from /login. Fix only these files."
Claude Code respects .gitignore, but sometimes you
need finer control — blocking large generated JSON files, compiled bundles, or
minified assets that Git tracks but Claude shouldn't read.
Create a .claudeignore in your project
root. Same syntax as .gitignore. Directly
reduces token usage and eliminates noise from every response.
Chapter 4
Advanced Workflows
Structure every non-trivial task using Scope → Scaffold → Implement → Verify. Tap any step to reveal the exact prompt pattern.
Chapter 5
Pro Configuration
Steer Claude Code's default behaviour per project with a
CLAUDE.md
file in your project root. Define conventions and constraints once — never repeat them in
prompts.
# Project Rules for Claude Code
## Stack
- React 18 + TypeScript (strict) + Tailwind CSS
- State: Zustand. Data fetching: React Query.
## Constraints
1. Functional components and hooks ONLY.
2. TypeScript strict — never use `any`. Prefer `unknown` + type guards.
3. CSS via Tailwind only — do not generate .css or .scss files.
4. Before adding a dependency, state its purpose and ask for approval.
5. Every utility function must have a Jest unit test.
## Code Style
- Named exports only (no default exports except pages).
- Keep files under 250 lines — extract if larger.
- Errors must be handled explicitly; never swallow exceptions.
## Architecture
- Feature folders: src/features/<name>/{components,hooks,types,utils}
- Shared logic in src/shared/ only if used by 3+ features.
⚠️ Security Note
Never put API keys, secrets, or credentials in CLAUDE.md. Claude Code may include fragments of its instructions in error outputs. Keep instructions focused strictly on conventions and architectural constraints.