Step 8 of 14

Claude Code

Claude Code is your AI coding assistant. It lives inside the terminal and can read files, write code, run commands, and explain anything you ask. You never use it alone — you always launch it from Agent Deck.

Always speak to Claude in English. AI models understand English best, and all documentation, code, and error messages are in English. Even if you think in Italian, write your prompts in English.

How to Launch

Claude Code is not an app you download or a website you visit. It runs on the VPS, inside Agent Deck.

  1. SSH into the VPS from your Mac Terminal: ssh giorgia@207.180.228.169
  2. Agent Deck appears. Use ↑ ↓ to navigate to the Work group.
  3. Press Enter, then select the personal session (or the session for your project).
  4. Press Enter again. You're now in the terminal for that session.
  5. Claude Code launches automatically when you open the session. You'll see it ready for your instructions.
Never launch Claude Code from a random terminal. Always go through Agent Deck first. If you're not inside an Agent Deck session, Claude Code won't be in the right project folder and things will go wrong.

Basic Interaction

Once Claude Code is running, you talk to it by typing messages. Think of it like a chat — but instead of casual conversation, you give it instructions and it does work.

You say Claude does
"Read the file src/main.py and explain what it does" Opens the file, reads the code, and gives you a plain-language explanation
"Create a new file called config.json with these settings..." Creates the file with the content you described
"Fix the bug on line 42 of server.py" Reads the file, identifies the issue, and proposes a fix
"Run the tests and tell me if they pass" Executes the test command and reports results
Be specific. Instead of "fix the code", say "fix the error in the login function that happens when the password is empty." The more detail you give, the better the result. See Prompt Building for how to write great instructions.

Plan Mode

Sometimes you want Claude to think before it acts. Plan Mode lets you review what Claude intends to do before it makes any changes.

  1. Type /plan to enter Plan Mode. Claude will now plan instead of execute.
  2. Give your instruction: "Refactor the database module to use connection pooling."
  3. Claude writes a plan — what files it will change, what it will add, what it will remove.
  4. Read the plan. If it looks good, say "go ahead" or "execute this plan."
  5. If you disagree, say "no, instead do X" and Claude revises.
When to use Plan Mode: Use it for big changes — refactoring, adding new features, restructuring files. For small tasks like "add a comment to line 10", just ask directly.

The Iterative Workflow

Working with Claude is not one-and-done. The best results come from a back-and-forth conversation:

  1. Ask — Give Claude a clear instruction.
  2. Review — Read what Claude did. Check the output, the code, the explanation.
  3. Refine — If it's not right, tell Claude what to change. "Good, but make the error messages more descriptive" or "This is close, but use a dictionary instead of a list."
  4. Repeat — Keep refining until you're satisfied.
Bad workflow Good workflow
Ask once, accept whatever comes out Ask, review, refine, ask again until it's right
"Build me a complete application" "First, create the data model. Let me review it. Now add the API. Let me check..."
Give up if the first result isn't perfect Say "that's not what I meant, I want X instead"
Don't be shy about saying "no". Claude is not offended. If the result is wrong, say so clearly: "This is wrong because X. Instead, do Y." Being direct gets better results.

Session Management

Claude Code sessions use memory and context. Here's how to manage them:

Command What it does When to use it
/exit Closes Claude Code completely When you're done working and want to go back to the terminal
/clear Clears the conversation history but keeps Claude running When you're switching to a different topic and the old context is getting in the way
When to clear: If Claude starts giving confused or irrelevant answers, it might be because the conversation got too long. Use /clear to start fresh, then re-explain what you need.

Next: Learn how to write great prompts in the Prompt Building guide — this is the most important skill for working with any AI.

← Agent Deck Prompt Building →