Claude Code guide: 10 habits and 2026 pricing

Three in the morning, and you have stared at the same file for three hours while an 80-line stack trace refuses to say where the bug lives. You paste the whole trace into the terminal with four words, "diagnose root cause first", and thirty seconds later Claude has found an off-by-one in another file and fixed it. This Claude Code guide collects what I wish I had known in week one: what the tool is, what it costs in euros and the ten habits that pay it back. The tests were green by half past three.

What is Claude Code, and how is it different?

Claude Code is an agentic coding tool from Anthropic, meaning a program that reads your codebase, edits files, runs commands and keeps working through a task instead of suggesting the next line.¹ It runs in the terminal, in VS Code and JetBrains, in a desktop app and in the browser. Your project settings follow you between them.

Three things together change how the work feels. Claude runs your own tools such as npm, pytest and git, so you stop copying commands between windows, and it can commit and open pull requests when you ask. A file called CLAUDE.md in the project root carries your coding standards and architecture decisions, and Claude reads it at the start of every session.¹ The third piece is the agent loop itself, which plans, edits, checks and tries again until the check passes.

A normal Thursday with Claude Code

The ticket says "add password reset to the auth flow", so you open the terminal, run claude, switch to plan mode and ask for the feature using the existing login code as the pattern. Claude reads the login file and returns a plan with three files to change, two new ones and a database migration, which you adjust in one place and approve.

After that it writes the code while you commit every few steps, and at the end you ask for a diff with one sentence per change. Two changes sit in files you never mentioned, so you tell it to roll those two back and keep the rest. The feature took 45 minutes against the three hours of writing and one hour of review it used to take. Reading that last diff is still your job.

The 10 habits in this Claude Code guide

Each habit solves one concrete problem, and each has saved me at least a full working day at some point.

1. Plan mode before anything non-trivial

Claude sometimes starts coding on an assumption it never checked with you, and ten minutes later the wrong feature exists. Press Shift+Tab until the status bar shows plan mode on, and Claude reads files and answers questions without changing anything until you approve a plan.² Skip it only when you could describe the whole diff in one sentence.

2. Run /init and keep CLAUDE.md short

Repeating "we use TypeScript strict and never commit without tests" in every session wastes the first minutes of each one. Run /init to generate a starter CLAUDE.md from your project, then add the commands, style rules and gotchas Claude cannot guess from the code.² Keep the file short, because a bloated CLAUDE.md makes Claude ignore the rules that matter.

3. One task per conversation with /clear

A database refactor followed by a frontend bug in the same session fills the context window, meaning the working memory of the conversation, with material that no longer applies. Run /clear between unrelated tasks, and after two failed corrections on the same problem clear and write a better first prompt instead.²

4. Point at a reference file

Describing the code you want in words gives you something that is roughly seventy per cent right and thirty per cent generic. Point at a real file instead, for example "look at src/auth/login.ts and build password reset following exactly the same pattern". Claude then copies your naming and error handling.

5. Build in small steps

Asking for the whole feature in one go tends to produce something that works most of the time, and finding the missing part then takes longer than writing it yourself. Break it into schema, endpoint, validation, form and wiring, with a test after each step, so when something breaks you know exactly where.

6. Write the tests first

Tests written after the code tend to confirm what the code already does rather than what it should do. Describe the edge cases, ask Claude to write the tests first and then build until every test passes.² A check it can run lets you walk away.

7. Paste the full error and ask for the root cause

A summary of an error invites a quick patch of the symptom, while the full stack trace plus "diagnose root cause first" makes Claude reason before it fixes. Anthropic's own best practices use the same pattern, asking Claude to verify the build and address the root cause.²

8. Review the diff before you trust it

After a large change, ask "show a diff of every file you changed and explain each change in one sentence". Reading it takes thirty seconds and catches the edit Claude made for consistency in a file nobody asked it to touch. That edit is always there.

9. Checkpoints plus git

Every prompt creates a checkpoint, and pressing Esc twice or running /rewind restores the conversation, the code or both.² Checkpoints do not track changes made through shell commands, so commit before every larger change as well. Git is the undo that covers everything.

10. Recovery mode when you loop

Fixing A breaks B and fixing B breaks A, and three hours later the code is worse than it was at lunch. Tell Claude to stop, read the last working version from git and start over with a different approach, restating the goal in one line.² A clean session usually wins.

What does Claude Code cost in 2026?

Claude Code is included in Pro, Max, Team and Enterprise, while the Free plan does not include it.³ Anthropic prices in US dollars, so the euro figures below are rough conversions. We used 0.85 euros to the dollar.

Pro at about €17 a month pays for itself when Claude saves you thirty minutes once a week, which is a low bar for anyone who writes code for a living. Usage limits on Pro and Max are shared across Claude and Claude Code, so a busy day in the chat app eats into the same budget as the terminal work.⁴

Which Claude model should you pick in Claude Code?

Anthropic lists Opus 5.5 as the daily driver for agentic coding and enterprise work, and Sonnet 5 as the high-performance model for coding and agents.³ Our rule of thumb is Sonnet 5 for everyday edits and tests, and Opus 5.5 when you plan something large or get stuck in a loop. Your plan's limit decides the rest.

Setup in five minutes

# 1. Install (macOS, Linux, WSL)
curl -fsSL https://claude.ai/install.sh | bash

# 2. Start inside your project and log in with a paid plan
cd your-project && claude

# 3. Generate the context file
/init

# 4. Ask for something concrete
claude "write tests for the auth module, run them, and fix any failures"

On Windows the native installer runs in PowerShell with irm https://claude.ai/install.ps1 | iex, and Homebrew and WinGet installs exist too, although those two do not update themselves.¹

Where does your code go under GDPR?

Claude Code runs locally but sends your prompts, the files it reads and the model's output over an encrypted connection to the model vendor.⁵ Your code does not stay entirely on your laptop. Under commercial terms, meaning Team, Enterprise, the API and cloud platforms, Anthropic does not train models on code or prompts sent to Claude Code, and standard retention is 30 days. Consumer Pro and Max accounts follow a training setting each user controls, which is why a company should put developers on Team seats.⁵

A company that must keep processing inside Europe can point Claude Code at Claude on Amazon Bedrock, where EU inference profiles route requests only to EU Regions such as Frankfurt (eu-central-1).⁶ On Bedrock, telemetry and error reporting to Anthropic are off by default, which makes the data flow easier to explain to a data protection officer.⁵ The wider picture for a whole engineering team is in agentic AI with Claude Code.

Rolling Claude Code out across a team

Learning it alone is one thing, and getting fifteen developers to share one CLAUDE.md pattern and one review flow is another. Through satori-claude the licences arrive on one euro invoice for €490 once plus €14 per user per month, with the licence at €24 for Standard or €110 for Premium. Team training runs as a satori-workshop at €280 per user for two hours from eight users, or as satori-sprint at €400 per user. Each developer still practises alone. All satori. prices are fixed and exclude VAT.

Open the repository you work in most, run /init, and cut the generated CLAUDE.md down to the ten lines Claude could not guess from the code. That takes under fifteen minutes, and you end up with one file every session on that project reads first.

---

Sources

Footnotes

1. Anthropic (2026). Claude Code overview. Claude Code Docs. https://code.claude.com/docs/en/overview ↩ ↩² ↩³

2. Anthropic (2026). Best practices for Claude Code. Claude Code Docs. https://code.claude.com/docs/en/best-practices ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷

3. Anthropic (2026). Plans and Pricing. https://claude.com/pricing ↩ ↩²

4. Anthropic (2026). Using Claude Code with your Pro or Max plan. Claude Help Center. https://support.claude.com/en/articles/11145838-using-claude-code-with-your-pro-or-max-plan ↩

5. Anthropic (2026). Data usage. Claude Code Docs. https://code.claude.com/docs/en/data-usage ↩ ↩² ↩³

6. Amazon Web Services (2026). Supported Regions and models for inference profiles. Amazon Bedrock User Guide. https://docs.aws.amazon.com/bedrock/latest/userguide/inference-profiles-support.html ↩