Claude Code — Quick Start Guide

Add features to your existing repo — fast.

Published
Author Basil
claudeaidevtoolsguide

Launch in Your Repo

# Navigate to your project
cd /your/project

# Launch Claude Code
claude

# Or open in a specific directory
claude --add-dir ./src

First Session Tip: Ask Claude to orient itself first: “Read the README and key source files, then summarize the architecture so we can plan new features.” This primes Claude with codebase context before you start building.


Feature Development Workflow

  1. Describe the feature — Give context: what it does, which files are likely involved, and any constraints (framework, style, etc.)

  2. Let Claude plan first — Say “plan this before implementing” — Claude will outline the approach and you can course-correct early.

  3. Review diffs in your IDE — Changes appear as visual diffs in VS Code / Cursor / Windsurf. Accept or reject per file.

  4. Iterate in the same session — Stay in one session for a feature — Claude remembers all prior context and changes made.

  5. Commit with Claude — Say “commit this with a good message” — Claude writes the commit and runs git for you.


Slash Commands

CommandDescription
/clearReset conversation context (new feature = new session)
/compactSummarize history to free up context window
/modelSwitch between Sonnet 4.6 and Opus 4.6
/resumeResume a previous session by name or title
/reviewAsk Claude to review current changes
/permissionsManage what Claude can auto-approve
/doctorDiagnose config and permission issues
/bugReport a bug directly to Anthropic
/configEdit settings, release channel, model defaults
/memoryView/manage what Claude remembers across sessions

Keyboard Shortcuts

  • Alt+P — Switch model mid-prompt
  • Ctrl+S — Stash/restore current prompt
  • Ctrl+C — Cancel current task
  • Ctrl+C ×2 — Kill all background agents
  • Tab — Tab autocomplete (bash mode)
  • Ctrl+B — Background current task
  • Alt+Del — Delete word (bash mode)
  • Alt+→ — Navigate word (bash mode)

Power Prompts for Feature Dev

  • Explore the codebase and explain how [feature area] works before we add anything
  • Add [feature] to [file/component]. Follow the existing patterns and coding style
  • Plan the implementation of [feature] step by step, don’t write any code yet
  • Write tests for the changes you just made
  • Find all the places this change could break something and fix them
  • Refactor [file] to be cleaner without changing behavior, then explain what you changed
  • Create a new branch, implement [feature], and open a PR with a clear description
  • Check the open GitHub issues and fix any related to [area]

CLAUDE.md — Your Persistent Context File

A CLAUDE.md file in your project root is read by Claude at the start of every session. It’s your standing instructions — coding standards, architecture notes, constraints, and preferences that Claude should always know.

# CLAUDE.md example

## Stack
React 18 + TypeScript, Vite, Tailwind CSS
Backend: Node.js + Express, Supabase

## Code Rules
- Use functional components, no class components
- All new components go in src/components/[feature]/
- Always add JSDoc to exported functions
- Run npm run lint before committing

## Architecture
Multi-agent platform using Agno framework.
Agents communicate via WebSocket (see src/agents/).

Pro Tip: Ask Claude to generate your CLAUDE.md: “Read the codebase and write a CLAUDE.md with our stack, folder structure, and coding conventions.”


Best Practices

  • One feature per session — Use /clear between unrelated features. Context pollution causes Claude to make inconsistent decisions.

  • Plan before you build — Always say “plan first” for anything non-trivial. Catch architectural issues before they’re in code.

  • Reference specific files — Use @filename to pin Claude’s attention. “Look at @src/agents/frontend.ts and add X” beats vague descriptions.

  • Ask for tests — Append “and write tests for this” to any feature request. Claude knows what edge cases to cover.

  • Use Opus for complex work — Switch to Opus 4.6 via /model for multi-file refactors or architectural decisions. Sonnet for quick edits.

  • Set permissions upfront — Run /permissions to pre-approve safe ops (file edits, npm install). Reduces friction during long sessions.

  • Use /compact aggressively — For long sessions, run /compact before adding a new subtask. Keeps context fresh and responses accurate.

  • Iterate, don’t overspecify — Give Claude a direction and refine. Overspecified prompts often produce worse results than conversational iteration.


Useful CLI Flags

# Run a one-off task non-interactively (great for scripts)
claude -p "Fix all TypeScript errors in src/"

# Add extra directories to context
claude --add-dir ../shared-lib

# Force reinstall / bypass lock issues
claude install --force

# Update to latest
claude update

# List configured agents
claude agents