Claude Code Cheatsheet - Everything you need to Vibe Code

Claude Code Cheatsheet - Everything you need to Vibe Code

Claude Code is a terminal-first AI assistant built to slot into your existing dev workflow - not replace it. It reads your codebase, runs commands, writes and edits files, searches the web, and even spins up sub-agents for complex jobs. Once you understand how it thinks, it becomes a genuine multiplier on your daily output.

Installation & First Run

Install it once, globally:

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

Then just type claude in any project folder to start.

Vibe code from anywhere - install Claude Code on a remote server and SSH in from any device. No laptop battery drain, no local resource limits, persistent sessions around the clock. Order a cloud server at DCXV and you're up in minutes.

For remote use, pair it with tmux so your Claude session survives disconnects:

# Install tmux (Debian/Ubuntu)
apt install tmux

# Start a named session
tmux new -s claude

# Detach - session keeps running after you disconnect
# Press: Ctrl+B, then D

# Re-attach from any device
tmux attach -t claude

# List running sessions
tmux ls

# Kill a session when done
tmux kill-session -t claude

Initial Setup Commands

A handful of slash commands will make your life easier right away:

• /permissions → Choose which tools Claude can use without asking every time (Git, Bash, file ops, etc.)
• /terminal-setup → Enables Shift+Enter for multi-line prompts - highly recommended
• /theme → Pick a color scheme (dark, light, or Daltonize for color blindness)
• /install-github-app → Connect to GitHub so you can tag @claude on issues and PRs
• /config → General settings, including notifications
• /memory → View and edit what Claude has remembered across sessions
• /model → Switch between Claude models mid-session
• /usage → Show token usage and estimated cost for the current session
• /clear → Wipe the current session context and start fresh
• /review → Trigger a code review of recent changes

If you're on macOS, enabling system dictation lets you speak longer prompts instead of typing them.

Codebase Q&A - Where to Start

The easiest way to get value on day one is asking questions about your codebase. Claude explores the code locally to answer - no upload needed, no context limit workarounds.

Some prompts that work well:

- "How is @RoutingController.py used?"
- "How do I add a new @app/services/ValidationTemplateFactory?"
- "Why does recoverFromException take so many arguments? Check git history."
- "Why did we fix issue #18383 with that if/else in @src/login.ts?"
- "What did I ship last week?" - Claude will scan your git log
- "Summarize what @src/auth/ does in plain English."
- "Find all places where we call the payments API."
- "Which files changed the most in the last 30 days?"
- "What dependencies does @package.json have that might be outdated?"

Start with Q&A to get a feel for what Claude understands immediately versus what needs more context from you.

Built-in Tools

Claude Code ships with around a dozen tools out of the box:

• Bash - run any shell command
• File search, list, read, write - full filesystem access
• Web fetch & search - pull in external docs or search results
• TODOs - lightweight task tracking
• Sub-agents - delegate complex multi-step work to parallel agents

Tip: teach Claude your team's custom CLI tools. For example:

Use the barley CLI to check error logs from the last training run. Use -h if you're unsure about flags.

Practical Editing Prompts

- "Propose a few fixes for issue #8732, then implement the one I pick."
- "Find edge cases not covered in @app/tests/signupTest.ts and update the tests. think hard."
- "commit, push, pr" - Claude learns this shorthand for standard git workflow
- "Use 3 parallel agents to brainstorm ways to clean up @services/aggregator/feed_service.cpp."
- "Refactor @src/utils.js to reduce duplication. Don't change the public API."
- "Add JSDoc comments to all exported functions in @src/api.ts."
- "Run the tests and fix any failures you find."
- "Look at the error in the logs below and find the root cause." - then paste the logs directly

Workflow Patterns

Different tasks call for different approaches:

• Explore → Plan → Confirm → Code → Commit - good for complex changes where you want oversight before anything is written
• Write Tests → Commit → Code → Iterate → Commit - TDD approach that keeps you anchored to expected behavior
• Write Code → Screenshot → Iterate - for UI work with Puppeteer or a simulator; Claude can react to visual output if given a way to check its work

Context Files (CLAUDE.md)

The more context Claude has, the better it performs. CLAUDE.md files are automatically read at startup - no need to re-explain your project every session.

Four places you can put them:

- /<enterprise-root>/CLAUDE.md    - org-wide policies, shared across all projects
- ~/.claude/CLAUDE.md             - your personal global context, applies everywhere
- <project-root>/CLAUDE.md        - project-specific, commit this to Git
- <project-root>/CLAUDE.local.md  - local overrides, not committed (for secrets or temporary notes)

Keep your CLAUDE.md concise. A bloated file slows Claude down and dilutes the signal.

Keyboard Shortcuts

• Shift+Tab - auto-accept file edits (Bash commands still need manual approval)
• # - save a memory to the relevant CLAUDE.md file
• ! - drop into Bash mode; the command and output go into context
• @ - mention a file or folder to pull it into the current session
• Esc - cancel whatever Claude is doing
• Double-Esc - jump back in history (combine with --resume to continue a session)
• Ctrl+R - verbose mode: shows Claude's reasoning and which tools it's considering

Scripting with the SDK

For automation, CI pipelines, or non-interactive jobs, the CLI doubles as an SDK:

claude -p "what did I do this week?" \
  --allowedTools Bash(git log:*) \
  --output-format json

It's also Unix-friendly - pipe in and out:

git status | claude -p "summarize my changes" --output-format=json | jq '.result'

Useful CLI flags:

--max-turns N        - limit the number of agentic steps
--continue           - pick up from where the last session left off
--resume <session-id>  - resume a specific past session
--no-cache           - skip prompt caching, useful for benchmarking

Running Multiple Sessions

Power users run several Claude instances at once - one per repo, one per task, or one per branch:

• Multiple terminal tabs with separate repo checkouts
• Git worktrees - a single checkout, multiple working branches
• SSH + tmux for remote multi-instance setups
• GitHub Actions to launch Claude jobs in parallel CI pipelines

Summary

Claude Code rewards investment. The more you configure it (CLAUDE.md, allowed tools, slash commands), the less friction you have per task. Start with codebase Q&A, build up your context files, learn the keyboard shortcuts, and eventually wire it into your CI - that's the full arc from beginner to power user.

Questions about AI tools for your infrastructure? Reach us at sales@dcxv.com