Mastering Claude Code in 30 Minutes

Some quick tips on using Claude Code:

Start with basic questions about the codebase. This will help you dial in what the tool is capable of and how to be prompt it.
When coding, instead of asking Claude code to immediately start, ask it to first form a plan.

Examples:
- "Propose a few fixes for <problem> then implement the one I pick."
- "Identify edge cases that are not covered in <test file>, then update the tests to cover these. Think hard."
- "Use 3 parallel agents to brainstorm ideas on how to clean up <file>."
Connect your team's tools. These are usually either shell tools or MCP tools.
Make use of well-established workflows. They're very powerful and useful. Some of these include:
- Explore → Plan → Confirm → Code → Commit: Figure out the root cause for <problem>, then propose a few fixes. Let me choose an approach before you code. ultrathink
- Write tests → Commit → Code → Iterate → Commit: Write tests for <test file>. (Note that the tests won't pass yet, since the feature has not yet been implemented. Then commit. Then update the code to make the tests pass. Then commit.)
- Write code → Screenshot Result → Iterate: Implement <design image>. Then screenshot it with Puppeteer and iterate until it looks like the mock.
The key here is that Claude needs some way to check its work so it can iterate. Give it a tool that it can use for feedback.
Use extended thinking mode. Claude supports this with four keywords: think, think hard, think harder and ultrathink. Each causes the model to think progressively harder about the problem.
Give Calude more context. The more context it has, the smarter it will be. This can be done in several ways:
- claude.md and claude.local.md: These files will get read into context automatically. You'd place common commands, common tools, architectural decisions and important files. Try to keep this short. You can also place these files in nested directories and they'll get pulled when Claude works in those directories.
- Slash commands: You can create your own commands. They're just markdown files.
- At-mentioning filenames: Using @ and then typing a file name will automatically pull it into context.
- MCP resources: You can define a .mcp.json file and it'll automatically download and install MCP servers.
It's worth your time to fine-tune the context. Use a prompt to tune it. This will improve performance of Claude dramatically.

This can all be checked in to version control, and can be changed at enterprise, user, project or local levels.
Make use of keybindings. There are a bunch of them built in to Claude. Below are some of the most helpful ones.
- Shift + Tab: Auto-accept edits so you don't have to okay every change
- #: Create a memory. You can tell Claude Code to remember something, like how to use a tool correctly, and it'll remember it from then on.
- !: Enter bash mode. This will run locally, but will also put the output into the context.
- @: Add a file or folder to the context
- Escape: Stop what Claude is doing. You can do this at any time.
- Escape + Escape: Jump back in history
- Control + R: Set the output to verbose, which allows you to see everything in Claude Code's context.
- /vibe: Enter vibe-coding mode.
Claude Code comes with an SDK that's scriptable. You can use the -p flag to pass a prompt. You can also use --allowedTools and --output-format for finer control. You can pipe into and out of the command.
You can launch multiple Claudes in parallel.
Claude Code is multi-modal. You can drag in an image, give it a file path, or give it a URL, and it can read and understand it.