Using with Claude Code

Claude Code is Anthropic's terminal-based AI coding agent. It reads your project files, runs commands, and writes code directly. This guide shows you how to get it fully loaded with your build package and building from minute one.

How Claude Code uses your package

Claude Code automatically reads any CLAUDE.md file in the project root at the start of every session. The CLAUDE.md in your package instructs Claude to then readproject.md and all files in /skills/ — giving it full context of your validated idea, tech stack, and implementation approach before you type a single message.

You never need to re-explain your app. Every new session starts with Claude fully briefed.

Installation

Install Claude Code globally via npm:

bash

npm install -g @anthropic-ai/claude-code

Then authenticate:

bash

claude

This opens a browser window to connect your Anthropic account. Once authenticated, Claude Code stores your credentials locally.

ℹ️

Requires an Anthropic accountClaude Code uses your Anthropic API usage. Make sure you have an active plan at console.anthropic.com.

Starting your first session

Navigate to your project root (where CLAUDE.md lives) and launch:

bash

cd my-app-name
claude

Claude Code will immediately read CLAUDE.md. Then paste this as your first message:

text

Read project.md and all files inside /skills/.

Once you've read everything, confirm you understand the project by telling me:
1. What this app does in one sentence
2. The full tech stack
3. Which skill domains are covered
4. What you'd suggest building first

Do not write any code yet.

💡

Why 'do not write any code yet'This forces Claude to confirm its understanding before acting. If the summary is wrong, correct it now — not after it's written 200 lines in the wrong direction.

Recommended build order

Once Claude confirms it understands the project, use this prompt to get a structured build plan:

text

Based on project.md and the skill files, give me a phased build plan.
Phase 1 should be a working skeleton I can see in the browser.
Phase 2 should be the core feature loop.
Phase 3 onwards is everything else.

List specific tasks per phase. Keep each task small enough 
to complete in one Claude Code session.

Then work through it one task at a time:

text

Let's start Phase 1, Task 1. Follow skill-core-app.md conventions.

Keeping Claude on track

Between sessions, Claude Code re-reads CLAUDE.md automatically. But if it starts drifting — ignoring your conventions, using different libraries, making up its own patterns — reset it:

text

Stop. Re-read project.md and all /skills/ files.
Then tell me what you were doing wrong and how you'll fix it.

Useful session patterns

Starting a new feature

text

I want to build [feature name]. 
Before writing anything, check project.md and the relevant skill file.
Tell me your implementation plan first.

Debugging something broken

text

[paste error]

Check the relevant skill file before suggesting a fix. 
Don't change anything else while fixing this.

Code review before committing

text

Review what you just wrote against the conventions in skill-core-app.md.
List anything that doesn't match and fix it.

Tips for best results

One task per session — Claude Code works best with a single focused goal. Don't ask it to build three features at once.
Reference skill files explicitly — say "follow skill-auth.md" when working on auth. It reinforces which file governs that domain.
Commit after each working feature — Claude Code can and will make large changes. Commit often so you have rollback points.
Keep skill files updated — if you change a tech decision (e.g. switch from Resend to SendGrid), update the relevant skill file so Claude Code picks it up next session.

← PreviousSkill files Next →Cursor / Windsurf