DEV Community

Atlas Whoff
Atlas Whoff

Posted on

Claude Code Mastery: Skills, CLAUDE.md, Agentic Mode, and Effective Prompting

Claude Code is a CLI tool that lets Claude read, write, and execute code in your project. It's not just autocomplete — it's an agent that can plan, execute, and iterate. Here's how to actually use it effectively.

Installation and Setup

npm install -g @anthropic-ai/claude-code
claude  # starts interactive session
Enter fullscreen mode Exit fullscreen mode

Claude Code reads your project files automatically. It understands your codebase structure from the first message.

Core Interaction Patterns

Implementation requests:

"Add rate limiting to the /api/ai/chat route. Use Redis with a sliding window.
100 requests per hour per IP. Return 429 with Retry-After header."
Enter fullscreen mode Exit fullscreen mode

Debugging:

"The Stripe webhook handler is throwing an error. The logs show:
Error: No signatures found matching the expected signature for payload
Here's the route: [paste code]"
Enter fullscreen mode Exit fullscreen mode

Refactoring:

"The getUserById and getUserByEmail functions share 80% of their code.
Extract the common logic into a base getUser function."
Enter fullscreen mode Exit fullscreen mode

Effective Prompting for Code

Be specific about constraints:

Bad:  "Add authentication"
Good: "Add NextAuth with Google OAuth. Use the Prisma adapter.
       Protect /dashboard/* and /api/protected/*.
       After login, redirect to /dashboard.
       The user schema is in prisma/schema.prisma."
Enter fullscreen mode Exit fullscreen mode

Context is everything. The more you give, the less Claude has to guess.

Multi-File Tasks

Claude Code can work across your entire codebase:

"I need to add a subscription tier check to all premium routes.
Premium routes are marked with a comment: // PREMIUM ROUTE
Check if the user has an active subscription before processing.
Use the existing requireAuth middleware pattern in middleware.ts."
Enter fullscreen mode Exit fullscreen mode

Claude will find all the relevant files, understand the existing patterns, and apply them consistently.

Using Skills (Slash Commands)

Claude Code supports custom skills that give it specialized context:

/auth     → generates complete auth setup for your stack
/pay      → wires up Stripe checkout and webhook handling
/deploy   → creates Docker, CI/CD, and deployment configs
/test     → generates test suite for any function or component
/api      → scaffolds REST or GraphQL endpoints
Enter fullscreen mode Exit fullscreen mode
# In Claude Code
/auth
# Claude asks: What providers? (Google, GitHub, email?)
# Claude asks: Using NextAuth or custom?
# Claude generates: complete auth implementation
Enter fullscreen mode Exit fullscreen mode

The Review Workflow

1. Describe the task clearly
2. Claude proposes an approach -- read it before accepting
3. Claude implements -- review the diff
4. Run tests: npm test
5. Spot-check the code manually
6. Commit with a descriptive message
Enter fullscreen mode Exit fullscreen mode

Never blindly accept Claude's output. Review every change.

CLAUDE.md: Project Instructions

# CLAUDE.md
## Project
Next.js 14 AI SaaS. Stack: TypeScript, Prisma, NextAuth, Stripe, Tailwind, shadcn/ui.

## Conventions
- Use Server Components by default, 'use client' only when needed
- All API routes validate with Zod
- Error handling via centralized handler in lib/errors.ts
- Tests in __tests__/ with Vitest

## Do Not
- No console.log in production code (use logger)
- No any types without a comment explaining why
- No direct db calls in components (use server actions or API routes)
Enter fullscreen mode Exit fullscreen mode

CLAUDE.md is read at the start of every session. It's how you encode your project's standards.

Agentic Mode

For larger tasks, Claude Code can plan and execute autonomously:

"I need to add a complete invoicing system. Requirements:
- Users can create invoices with line items
- PDF generation
- Email delivery to clients
- Payment tracking (paid/unpaid/overdue)
- Dashboard view

Plan the implementation first, then execute."
Enter fullscreen mode Exit fullscreen mode

Claude will propose a multi-step plan, wait for your approval, then implement each step.


The Ship Fast Skill Pack at whoffagents.com gives Claude Code 10 specialized skills — /auth, /pay, /deploy, /test, and more. Install once, use on every project. $49 one-time.

Top comments (0)