Curated by aiwithpravin — the exact steps I use to run Claude Code safely on a real codebase, from first install to your first AI-assisted fix.

What Is Claude Code

Claude Code is an AI coding agent that runs in your terminal. It reads your repo, plans a change, edits files, runs your tests, and fixes its own mistakes — you review the diff instead of writing from scratch.

Before You Start

• A codebase in git (so you can always undo)
• A terminal on your machine (Mac, Linux, or Windows with WSL)
• An Anthropic account with Claude Code access

Step 1: Point It At Your Repo

Open your project folder in the terminal, launch Claude Code, and give it one plain-English task describing what you want changed. No special syntax needed.

Step 2: Let It Plan Before It Touches Anything

Claude Code reads the relevant files first and explains its plan before editing anything. Read this plan closely — if it’s reading the wrong files or misreading the task, stop and clarify before it makes any changes.

Step 3: It Edits, Runs, and Self-Corrects

It edits the code, runs your tests, and if something fails, it reads the error and fixes its own mistake — repeating this loop until the tests pass.

Step 4: You Review, Not Rewrite

Your job shifts from typing code to reviewing a diff. Read every changed line before merging — this is the step people skip, and the one that saves you later.

3 Rules Before You Trust It

1. Always work on a separate branch — never commit straight to main.
2. Always read the full diff before merging, even for small changes.
3. Never point it at production data or real secrets on a first run.

Prompts That Work Well