Claude Code: TL;DR Playbook

• 2 min read • Systems Engineering

A single-page cheatsheet for making the most of Claude Code in daily development

When I first started using Claude Code, I tried to learn and understand best practices. Here’s the one‑screen cheat sheet I prepared for myself to get the most out of it. For more details read the official documentation and for a deep dive follow this article - most of the TLDR items are based on it.

  • Create CLAUDE.md. Keep only critical context; Claude auto‑attaches it to every session. There’s no one‑size template, just add whatever matters for your project.
  • Keep a /docs folder. Store living project docs there; let Claude update and reference them as you work. It helps with context and memory management, and leaves you with a proper knowledge base you can consult later
  • Use MCPs. Personally, I’ve found Context 7, Puppeteer, Playwright, and GitHub helpful for my development. I could be different for you. You can browse this catalog for more options.
  • Parallelize Claude’s work (detailed guide).
    • Offload tasks to sub‑agents to move faster and preserve main session’s context.
    • When working on multiple tasks in the same project, use git worktree to clone a separate branch into a new directory, handle the task there, and keep a separate Claude session for that directory. It helps to multitask with no context clashes.
  • Proactively manage your context. Treat tokens like runway: end each work phase by running /compact or /clear, offload side tasks to sub‑agents, and save large outputs in /docs. Proactively managing your context prevents Claude from unexpectedly hitting the token limit, force‑compact in the middle of a task, and forgetting finer details.
  • Analyze and plan first. Use Shift+Tab to toggle Plan mode. Deepen analysis with trigger words think < think hard < think harder < ultrathink.
  • Useful slash commands.
    • /init - analyze project and create CLAUDE.md
    • /resume - continue a previous session
    • /agents - manage sub‑agents
    • /mcp - manage MCPs
    • /clear - clear current context
    • /compact - summarize session to free up context space
    • /review - do a pull‑request review
  • Create your own commands. Personally, I’ve created a helper command that generates commit messages when I commit manually
  • Run Claude headless (-p).
    • Local scripts (npm or Bash)
    • CI/CD pipelines
    • Unix chaining: claude -p "cmd" | next_tool
  • Paste images (Ctrl + V). Screenshots drop straight in conversation.
  • Allow safe commands. Let Claude run low‑risk commands without asking you.
  • Official GitHub integration. Use Claude to auto‑review PRs, comment on issues, fix bugs in your repository
View All Notes