Detailed content
📄 CLAUDE.md (Memory Files)
O CLAUDE.md is the file Claude reads in every session to get to know your project. It's not flashy, no one picks it as a favorite—but it's the groundwork, the bare minimum. Without it configured, the experience simply isn't good: Claude starts from scratch every day, unaware of your rules, your stack, or how you work.
🧩 What to put in the file
O CLAUDE.md is plain text. It carries the context you don’t want to repeat every session:
- •Project stack, conventions, and commands (how to run it, how to test it).
- •Style rules and preferences for how you want it to work.
- •It can be global (applies to everything) or project-specific (at the repo root).
💡 Practical tip
Start small: three or four lines can already change the experience. Refine them as you notice Claude making the same mistake — each recurring correction is a candidate for becoming a line in the CLAUDE.md.
Read every time it opens.
The bare essentials.
Everything or one project.
Grows with use.
♻️ Context Lifecycle
Three commands to manage the conversation: /clear clears and starts from scratch, /compact compacts what’s already been said and continues, and /resume resumes a previous session. Basic features — but you use them every day because they control the session’s scarcest resource: context.
/clear — clear
Clears the window and starts a new conversation. Use it when you change topics and don’t want to carry the old context along.
/compact — compact
Summarizes what has happened so far to free up space and continues the same task. Ideal when the context window is filling up but you’re not done.
/resume — resume
Return to a previous session and pick up where you left off. Good for tasks that span several days.
Clear the conversation.
Summarize and continue.
Resumes a session.
Useful every day.
🔎 Web Search & Fetch
Claude has native web search that cites sources — so you can verify where each piece of information came from. And web fetch launches a search on the spot from a link. Basic, but still useful: it brings you current information without leaving the terminal.
✓ When to use it
- ✓Check a recent fact that may be outside the model's knowledge.
- ✓Find and cite sources in an answer that needs to be verifiable.
- ✓Pull content from a specific URL you already have.
✗ Don’t confuse
- ✗Search scans the web; fetch starts from a specific link.
- ✗It's not an internal code search tool — there are other tools for that.
- ✗No citation isn't the same thing: the differentiator here is citing sources.
💡 Practical tip
When asking for something that depends on current information, explicitly ask Claude to search and cite. You gain traceability and reduce the risk of outdated responses.
Verifiable response.
Without leaving the terminal.
Part of a URL.
Recent info, in the moment.
🔐 Permissions
Permissions controls what Claude can do without asking: which commands to run, which files to edit, which tools to use. It may seem minor, but it’s the foundation of Auto Mode (Tier A): getting permissions right now is what unlocks autonomous workflows later.
🛡️ The three modes
- •Allow — Claude does it without asking.
- •Ask — it asks for confirmation before acting.
- •Deny — permanently blocks that action.
allow: read commands you trust ask: file edits and commands that change state deny: destructive actions that should never run on their own
💡 Practical tip
Start restrictive and gradually allow what you trust. Every permission you grant carefully today means less friction tomorrow — and it’s the natural path to Auto Mode.
Without asking.
Asks for confirmation.
Blocks it permanently.
Unlocks Tier A.
📎 File Uploads & Output Styles
Two simple things that make a big difference in your daily comfort: drop files (images, PDFs, documents) directly in the conversation to provide rich context; and output styles, which change how Claude formats responses to suit your preferences.
✓ File Uploads
- ✓Drop in an image so Claude can see what you see.
- ✓Send PDFs and docs instead of copying and pasting excerpts.
- ✓Rich input = more precise, context-aware answers.
✓ Output Styles
- ✓Change how Claude formats its responses.
- ✓More concise, more detailed, more technical—your way.
- ✓Consistent output without repeating the instruction every time.
💡 Practical tip
Before describing a screen error in three paragraphs, drop in the screenshot. And if you always ask for the same response format, configure an output style and stop repeating yourself.
Visual context.
No copying and pasting.
Output your way.
Less repetition.
🧰 The rest of the foundation
A package of small plumbing pieces. None is a star on its own, but together they make the day more comfortable — shortcuts, visual comfort, and access from anywhere.
Speed & appearance
- •Fast Mode — Opus with faster output (/fast).
- •Custom Themes — /theme: auto, dark, colorblind-friendly, and custom themes.
- •IDE Extension — Claude Code in VS Code/JetBrains; good for getting started and for copying and pasting without the terminal’s weird wrapping.
- •Mobile Apps — iOS and Android.
Shortcuts & modes
- •Code Review — /code-review from the git diff.
- •Sandboxed Bash e Headless Mode.
- •Transcript Viewer (Ctrl+O) e History Search (Ctrl+R).
- •Shell Mode (prefix !) e Voice Dictation (/voice).
Ctrl+O transcript viewer Ctrl+R history search ! shell mode (roda comando direto) /fast saída mais rápida do Opus /theme troca o tema visual /voice ditado por voz
💡 Practical tip
Don't try to memorize everything at once. Adopt one or two per week — Ctrl+R and the prefix ! tend to become habits the fastest.
Faster Opus.
VS Code/JetBrains.
Ctrl+O, Ctrl+R, !
Desktop and mobile.
🏁 Module summary
Next: Module 1.2
With the foundation in place, move up a level to Tier C: Cowork, voice, and the web — good features that are rarely used.