Claude Code tips I actually use

A free 17-page PDF: 24 habits, shortcuts and settings that changed how I work with Claude Code, each with the lines to copy. My own mods and tools are in the back.

Download the guide (PDF, 17 pages, 6.7 MB)

Most of what makes Claude Code good is not a plugin. It is a handful of habits: a short AGENTS.md, a plan before a big change, a way for Claude to check its own work, a clean context. This guide is the 24 I use every day, in plain words, each with the line to copy. The add-ons I built are still in it, at the back.

What’s in it

01 Set it up once. A short AGENTS.md that every agent reads, auto memory, the four modes, permission rules, and the small settings worth changing: Shift+Enter for a new line, compact at 250k, a ping when Claude needs you, and a status line.

02 Every session. Plan first, then build. Give Claude a way to check its work. Show it the file, the screenshot or the log. Run a shell command with !, ask a side question with /btw, rewind when it goes wrong, and clear between tasks.

03 Bigger jobs. Subagents, skills and hooks, and which one you need when. A CLI before MCP. Sessions side by side, each in its own worktree, and claude -p for scripts, CI and runs over many files.

04 Safe and cheap. Model and effort, what a session costs, keys out of the chat, /sandbox, a second review before you push, and what to do when something feels off.

05 Cheat sheet. Every key, command and file from the tips on one page.

Appendix. My four mods (Clawd Cartoons, Clawd Park, Scenes, Secrets), my Azure plugin, Agent Office, and two tools by Kun Chen. Mods are not sandboxed, so the appendix starts with how to check one before you install it.

Short on time? Do these five

Give it a way to check its work (tip 07).
Plan first on anything bigger than a typo (tip 06).
Clear between tasks (tip 12).
Keep AGENTS.md short (tip 01).
Stop early, rewind freely (tip 11).

Let Claude pick the rest

The last page has a prompt you send to Claude Code together with the PDF. It looks at your setup (version, AGENTS.md, permission rules, hooks, MCP servers), asks what you use it for and what annoys you, and names the five tips that would help you most. It changes nothing until you say go.

I checked every tip against the Claude Code docs and on Claude Code 2.1.296. Claude Code moves fast. Found one that is out of date? Tell me on X and I will fix the guide.

Sources

Claude Code docs: best practices, memory, permission modes, permissions, hooks, subagents, skills, background sessions, headless, costs, mods. My repos: github.com/falkoro. Kun Chen: axi.md, no-mistakes. My video on compacting at 250k.