Practical Claude Code habits that save time, tokens, and mistakes
Short answer
A free guide to 20 Claude Code tips as of October 2026: CLAUDE.md, plan mode, context control, rewinding, subagents, skills, hooks, MCP, plugins and mods, permissions, headless mode, worktrees, models and effort, code review, and safe habits.
Claude Code is Anthropic’s agentic coding tool. It runs in the terminal, the desktop app, the web at claude.ai/code, and IDE extensions for VS Code and JetBrains. It reads your project, edits files, and runs commands with your permission.
Claude Code is fast. Good habits make it reliable.
02
Set up the project
1 · Run /init
It creates a CLAUDE.md file describing your project, commands, and conventions. Claude reads it at the start of every session.
2 · Keep CLAUDE.md short
Add build and test commands, style rules, and things Claude keeps getting wrong. Remove anything outdated.
3 · Point to files with @
Mention files with @ so Claude reads exactly what matters instead of searching.
4 · Paste screenshots
Paste an image of a design or an error. It is often faster than describing it.
03
Plan before building
5 · Use plan mode
Press Shift+Tab to cycle modes and pick plan mode. Claude researches and proposes a plan without editing files.
6 · Ask for tests first
For bigger changes, ask Claude to write or update tests, then make them pass.
7 · Give a clear finish line
Say what done looks like: “The page loads, tests pass, and nothing else changes.”
04
Control context and tokens
8 · Use /clear between tasks
A fresh context for each new task keeps answers focused and saves tokens.
9 · Use /compact on long sessions
It summarizes the conversation so you can keep going with less context.
10 · Pick the right model and effort
Use /model to switch. Use bigger models and higher effort for hard problems, and smaller ones, such as Haiku 5.5, for quick tasks.
05
Fix mistakes fast
11 · Press Esc to stop
Interrupt Claude as soon as it heads the wrong way, then redirect it.
12 · Rewind with checkpoints
Press Esc twice, or use /rewind, to go back to an earlier point in the conversation and code.
13 · Resume past sessions
Use /resume, or start with claude --continue, to pick up where you left off.
06
Extend Claude Code
14 · Use subagents
Subagents handle side tasks, such as searching or reviewing, in their own context. Since October 2026, you can set the effort level a subagent runs at.
15 · Turn repeat work into skills
Save instructions for tasks you repeat as skills or slash commands, so you never retype them.
16 · Add hooks
Hooks run your own commands at set moments, such as formatting code after every edit.
17 · Connect MCP servers and plugins
Connect tools such as GitHub, databases, or docs with MCP. Install plugins from marketplaces, and use mods, added in October 2026, to change how Claude Code itself behaves.
07
Work faster
18 · Set permissions once
Allow safe, frequent commands, such as running tests, so Claude stops asking. Keep risky commands on ask.
19 · Run in parallel with worktrees
Use separate git worktrees so several sessions can work on different tasks without clashing.
20 · Go headless
Run claude -p "your prompt" in scripts and automations for one-off tasks without the interactive screen.
08
Review before you ship
Read every diff before committing
Run tests and the app yourself
Use /code-review to ask Claude to check the changes for bugs
Never let it push to production without your check
09
Keep it safe
Never paste passwords, API keys, or tokens into prompts
Keep secrets in environment files that are not committed
Do not allow broad commands such as deleting files without asking
Use git, so every change can be undone
10
Master task prompt
Task: [what to build or fix].
Context: [files, links, or errors].
Rules: [style, libraries, what not to change].
Done means: [tests pass, behavior, output].
First read the relevant files and propose a short plan. Wait for my OK, then make the smallest change that works, run the tests, and summarize what changed.
The golden rule
Do not ask Claude Code to do everything at once. Give it context, a plan, and a clear finish line, then review what it built.
CLAUDE.md
Plan
Context
Rewind
Subagents
Skills
Review
Run /init in one project today and add your test command and three rules to CLAUDE.md.
It has 10 sections, starting with “Before you start”, “Set up the project”, “Plan before building” and “Control context and tokens”, plus 6 more sections.
How long does it take to read?
About 8 minutes. Every section is a short, self-contained card, so you can jump straight to the part you need.
Is this guide free?
Yes. The full guide is free to read on this page, with no sign-up.