Covers Claude (claude.ai and the desktop app) and Claude Code (terminal, VS Code and desktop).
Short answer: the Claude tips and tricks that matter most
- Limits: stay on one model per chat, keep CLAUDE.md under about 200 lines, and run /compact early. Switching models makes Claude re-read your whole chat at full price.
- Forgetful chats: use Projects, memory and a short CLAUDE.md. A rule in CLAUDE.md is a request. A hook is a guarantee.
- Done means done: use a stop hook, or give Claude a way to test its own work.
- Less clicking: use Research, connectors, Claude in Chrome, subagents, plugins and Auto mode where they fit.
- Automation: cloud sessions and routines keep working when your laptop is shut.
These Claude tips and tricks come with a short video for each tip, and answer the problems people actually type into Google: “claude limit reached”, “claude forgets context”, “claude says done but it isn’t”. Each tip names what it is, why it works, how to do it, and where it works (Claude chat, the Claude desktop app, or Claude Code). Features change fast, so check that a setting exists in your version before relying on it.
Contents
- Claude limit reached: 8 tips (1 to 8)
- Claude forgets context: 6 tips (9 to 14)
- Claude skills: 5 tips (15 to 19)
- Claude says it is done but it is not: 4 tips (20 to 23)
- Claude stops when you close your laptop: 2 tips (24, 25)
- Phone, notifications and planning: 5 tips (26 to 30)
- Research, Chrome, connectors, voice and more: 8 tips (31 to 38)
- Plans, slash commands, files and artifacts: 4 tips (39 to 42)
- Official docs power habits: 10 tips (43 to 52)
- Where each feature works (table)
- FAQ
Claude Limit Reached? 8 Tips to Get More Usage
Claude token usage is invisible until you hit a limit. Every tip below lowers how much Claude has to read, re-read or retry.
1. Pick one model per job and stay on it
What it is: each Claude model keeps its own saved copy (prompt cache) of your conversation. Why it matters: when you switch models with /model, the next request re-reads the whole chat with no cache hits, which costs more. How to fix it: stay on one model per chat. For an easy question, lower the effort level instead of switching. On Opus 5.5, Sonnet 5.5, Haiku 5.5 and Fable 5.1, changing effort keeps your cache.
2. Cheaper per token is not cheaper per job
A smaller model can need more attempts and more edits to reach the same result, and every retry costs tokens and your time. Start hard jobs on a stronger model and save the small model for simple, well defined work.
3. Use /advisor to mix models without breaking the cache
What it is: /advisor lets a cheaper model do the main work while a stronger model reviews key decisions. Why it works: it does not switch the main model, so your cached conversation stays intact. Claude consults the advisor before committing to an approach, when the same error keeps returning, and before saying a task is done. It is marked experimental, so expect it to change.
4. Do not crank effort to maximum
Higher effort makes Claude think longer and uses your limit faster. Leave the default effort alone until answers actually fall short, then raise it for that one job. In Claude Code you can change effort without losing your cache (see tip 1).
ultrathink in that single prompt. It asks for deeper reasoning on that turn only and leaves your effort setting alone. Phrases like “think hard” are just ordinary text, so use the effort setting for the real control. On Opus 5.5, Sonnet 5.5, Haiku 5.5 and the Fable models, thinking cannot be turned off.5. Keep CLAUDE.md under about 200 lines
What it is: CLAUDE.md is a plain text file Claude Code reads at the start of every session. Why it matters: the first 200 lines (up to 25 KB) load every time, and longer files mean more context and weaker adherence. How to fix it: keep only rules that apply to everything, and move part-specific rules into separate path-scoped rule files.
/init and Claude drafts a starting CLAUDE.md from your project, then you mostly delete lines. Do not describe your file layout, because Claude reads the repo itself. If a line keeps being ignored, adding IMPORTANT rarely helps. Either the file is too long and the line gets lost, or the rule cannot have exceptions and belongs in a hook. For rules about one part of the project, create a file in .claude/rules/ with a paths: filter so it loads only when Claude works there.6. See where your tokens go with /context
Type /context in Claude Code and it shows what is filling your window. You will usually find old tool output and an oversized instruction file. Use /compact to shrink the conversation.
/usage now shows which skills, subagents, plugins and MCP servers used your plan, plus recommendations such as “Trim your CLAUDE.md”. When the footer shows “f to fix”, press f to have Claude Code apply the ones it can. /insights writes a report on how you work and where you hit friction.7. Prefer a command-line tool over an MCP server
When a command-line tool exists for the job (GitHub, AWS, Google Workspace), it usually uses fewer tokens than the matching MCP server, because nothing is loaded into context until the command runs. A practical order: command-line tool first, API second, skill third, MCP server only when nothing else fits.
/mcp to see your configured servers and turn off the ones you are not using. MCP tool definitions are loaded on demand, so only their names sit in context until Claude uses one.8. Let Claude wait out the limit instead of losing the task
What it is: when a claude.ai usage limit is reached, Claude Code can wait in the open session and continue by itself after the reset. How it works: it is on by default in Claude Code v2.1.234 and later when you sign in with a claude.ai account, and the bottom lines show the reset time. Keep the session open.
/usage-credits opens your usage settings in the browser, where you can turn on usage credits to keep working past the limit.Claude Forgets Context? 6 Tips to Fix Long Chats
9. Put recurring context in a Project
What it is: a Project is a workspace in Claude with its own reference documents and instructions. How to use it: create a Project for work you return to, add your documents, and write a short instruction about who you are and how you want answers. Every chat in that project starts preloaded.
10. Turn on memory and past-chat search
Memory lets Claude carry your role, preferences and decisions across conversations. In Claude Code, each session starts fresh, so CLAUDE.md and auto memory carry context between sessions. Together with a project knowledge base, this removes most “start from zero” moments.
11. Run /compact early, not after the warning
A longer chat gives Claude more to sift through. Run /compact in Claude Code while the session is healthy, and start a fresh session for a new job instead of stretching one chat across unrelated tasks.
/compact a focus, for example /compact Focus on code samples and API usage. You can also add a “Compact instructions” section to CLAUDE.md. Claude Code also compacts on its own before the window fills, but /compact itself reads the whole conversation, so compacting a huge chat is a big request. /clear costs nothing: run /rename first, then /clear, and come back later with /resume.12. Tell Claude to update CLAUDE.md after every mistake
Say it plainly: “Update your CLAUDE.md so you don’t make that mistake again.” After a few weeks the file reflects how you work.
13. Rewind instead of arguing
When Claude goes down the wrong path in Claude Code, press Escape twice to rewind to an earlier message. The failed attempt leaves the context, instead of staying behind as noise that you keep correcting.
Esc once to stop Claude mid-action without losing work. /rewind (or Esc twice on an empty prompt) can roll back the code, the conversation, or both. If you have corrected the same thing twice and it is still wrong, run /clear and write one better prompt instead of a third correction. Checkpoints cover files Claude edited with its own file tools, not files changed only by a shell command, and they work alongside git.14. Ask side questions with /btw and try ideas with /branch
In Claude Code in the terminal, /btw lets you ask a quick question without adding clutter to the main conversation, and /branch forks the conversation at that point so you can test one approach, then try another. Command availability differs between the terminal and the desktop app, so check the slash menu in yours.
Claude Skills: 5 Tips to Stop Writing Useless Ones
15. What are Claude skills, in one line
A skill is a saved set of instructions that Claude reads and follows when a task matches. Only its short description loads each session, and the full content loads when the skill is used. Build it once (“weekly client report”) and you never have to explain the job again.
16. Write only the corrections
Claude already handles most routine steps. Try the job without a skill first, write down only what Claude got wrong, and put just that in the skill. A short skill is cheaper and easier for Claude to follow.
17. Test every skill against no skill
Run the same job with and without your skill and compare the results. If the skill does not clearly beat the plain output, trim it or delete it.
18. Let Claude draft the skill from a job you just corrected
After you finish a job and fix Claude’s mistakes along the way, ask Claude to turn those corrections into a skill. Because it captures only the places you corrected, the result stays small and focused.
19. Skills versus MCP, commands and subagents
- Skill: instructions for how to do something.
- MCP server or connector: a connection to an outside tool or data. Anthropic’s docs put it this way: MCP connects Claude to external services, and skills extend what Claude knows, including how to use those services.
- Command: a saved prompt you trigger with a slash.
- Subagent: a separate worker with its own context.
- Hook: an automatic check that always runs.
Claude Says It Is Done but It Is Not: 4 Tips
20. Use a stop hook so Claude must pass a check
What it is: a Claude Code hook is a rule that runs at a set moment every time. A stop hook fires when Claude tries to finish. How it works: a prompt hook makes one model call that answers yes or no with a reason. If the answer is no, the reason goes back to Claude and it keeps working. An agent hook starts a helper that can read files before deciding.
21. A rule in CLAUDE.md is a request. A hook is a guarantee
Claude follows CLAUDE.md as an instruction, not a guarantee. Anything that must happen every time belongs in a hook, because a hook fires on its event without Claude deciding. You do not need to write it by hand: describe the check to Claude and ask it to add the hook.
22. Make /goal provable
/goal keeps Claude working until a condition is met, but the checker reads the conversation, so “the report is finished” lets Claude simply claim it. Write the goal so proof appears in the chat: “the check script passes and you show me the output; stop after 15 turns.”
23. Let Claude test its own work
Give Claude a way to check results: a browser to open the page, screenshots to compare, a test to run. A verification loop is one of the biggest quality upgrades available, because Claude stops handing you unfinished work.
Claude Stops When I Close My Laptop: 2 Tips
24. Use cloud sessions and routines
What it is: a cloud session runs Claude Code on Anthropic’s infrastructure, so it keeps running after you close the laptop. How to start: claude.ai/code, the mobile app, the desktop app, or claude --cloud. You can steer it from any device, and it counts toward your plan limits with no separate machine charge. Scheduled tasks inside the desktop app need your computer awake, while routines run in the cloud. Only attach the connectors a routine really needs.
/loop polls inside the current session. Because a scheduled task cannot ask questions, say exactly what success looks like and what to do with the result.25. Run parallel jobs in separate worktrees
A worktree gives each job an isolated copy of your project on its own branch, so three or four jobs can run at once without overwriting each other’s files. Agent teams are a separate, experimental feature: a coordinated group of sessions, instead of helpers inside one session.
claude --worktree feature-auth, and run the same command with a different name in a second terminal. The repository needs at least one commit first.Phone, Notifications and Planning: 5 More Tips
26. Control Claude Code from your phone with Remote Control
Remote Control connects the Claude app on iOS or Android, or claude.ai/code, to a session running on your own computer. Claude keeps running locally, so your files and tools stay on your machine. Run claude remote-control in your terminal, or type /remote-control (or /rc) inside a session you already started, then open the session URL on your phone. It works on Pro, Max, Team and Enterprise plans, not with API keys. Your computer has to stay on. To keep a task going with the laptop closed, use a cloud session (tip 24).
27. Get a notification when Claude needs you
A Notification hook runs a command of your choice when Claude is waiting for permission or has been idle for about a minute. Add the hook to ~/.claude/settings.json, then type /hooks in the terminal to confirm it shows up. The Claude Code hooks guide includes ready-made commands for macOS, Linux and Windows.
28. Prompt hook or agent hook: pick the cheaper one that works
A prompt hook makes one model call and decides from the information it is given. An agent hook starts a subagent that can read files and run commands, for up to 50 turns (60 seconds by default). Use a prompt hook when the conversation is enough, and an agent hook when Claude must look at real files. Agent hooks are marked experimental.
29. Plan with Opus, build with Sonnet
The opusplan model setting uses Opus in plan mode and Sonnet to execute. Each toggle of plan mode is a model switch, so it starts a fresh cache (tip 1). Toggle plan mode on purpose, not back and forth.
Shift+Tab until the status bar shows “plan mode on”, or start with claude --permission-mode plan.30. Subagents start fresh. Forks keep the conversation
A normal subagent works in its own context window and does not see your chat. It starts from a short brief, does the job and returns a summary. A fork copies the conversation so far, so it already knows the context. Both count toward the same usage limits as your main chat, so more helpers means faster use of your plan.
Research, Chrome, Connectors, Voice and More: 8 Tips
31. Create a Claude Code subagent (the real steps)
What it is: a subagent is a helper that works in its own separate context window and returns only a short summary, which keeps test logs and search results out of your main chat. How to create one: (1) in Claude Code, type /agents to see your helpers; (2) tell Claude what you want, for example “create a subagent called code-reviewer that checks my code for bugs”; (3) Claude saves it as a markdown file in .claude/agents/ for one project, or ~/.claude/agents/ for all projects. You can also write that file by hand. The top of the file holds a name, a description (both required), the tools it may use, and a model (sonnet, opus, haiku, fable or inherit). A cheaper model suits simple jobs. How Claude picks one: by reading the description, or you can name the subagent yourself. Built-in subagents include Explore, Plan and general-purpose. Subagents can start their own helpers up to three layers deep, and up to twenty can run at once. They work in the terminal, in VS Code and in the desktop app.
model: haiku for simple jobs such as collecting files or checking links. Haiku costs much less per request. To use one right away, just say “use a subagent to investigate how our auth system handles token refresh”.32. Use Claude Auto mode to skip routine approvals
What it is: Auto mode is a permission mode in Claude Code where a separate safety classifier reviews each action before it runs, so you stop approving every step. How to switch it on: in the terminal, press Shift+Tab until Auto shows, or start with claude --permission-mode auto. In the desktop app and VS Code, pick Auto from the mode selector under the prompt box. What it blocks: force pushes, production deploys, and downloading and running unknown code. Edits in your working folder go straight through. It is not a guarantee, and after three blocks in a row or twenty in total it pauses and asks you again. Bypass permissions has no checks at all, so keep it for isolated containers.
autoMode.allow in plain English, and run /auto-mode-setup to have Claude draft the trusted environment entries from your project.33. Install Claude Code plugins
What it is: a plugin bundles skills, agents, hooks and MCP servers into one install. How to install: in the terminal, type /plugin and open the Discover tab, or install by name, such as /plugin install commit-commands@claude-plugins-official, then choose who gets it: just you, everyone on the project, or only this repo. For a plugin from another source, add its marketplace first with /plugin marketplace add owner/repo. In the desktop app, click the plus next to the prompt box and choose Plugins. In VS Code, type /plugins. A plugin can run hooks and servers on your machine, so read what it adds first, and remove plugins you stop using, because they can add to every message.
34. Use Claude Research for questions that need depth
What it is: Research mode runs many searches in a row, and each one decides what to look at next. How to use it: make sure web search is on, click the plus button at the bottom left of the chat, and choose Research. A blue indicator shows it is active. It works on the web, in the desktop app and on mobile, on Pro, Max, Team and Enterprise. It can also use connected tools like Gmail, Calendar and Google Docs, but only when you start it on the web or in the desktop app. It uses your limit faster than a normal chat, so save it for questions that really need depth.
35. Let Claude in Chrome do your browser chores
What it is: Claude in Chrome is a browser extension with a side panel where Claude can read pages, click, type, fill forms and work across tabs. How to set it up: open the Chrome Web Store, add Claude in Chrome, sign in with your Claude account, pin it from the puzzle piece menu, and click its icon to open the side panel. It starts in automatically approve mode, where Claude checks each action for safety and pauses to ask when it needs you. It does not buy things, create accounts or permanently delete files. It needs a paid plan and works only in Google Chrome. In the desktop app, turn it on under Customize, then Connectors. In Claude Code, type /chrome to check the connection.
36. Connect Gmail, Drive and other tools with connectors
What it is: a connector lets Claude read and use another service, such as Gmail or Google Drive, so you stop pasting text by hand. How to set it up: open Customize, then Connectors, click the plus, pick a service, click Connect and sign in. In any chat, click the plus button, hover over Connectors and switch on the ones you want for that chat. Claude only sees what your own account can see. On Team and Enterprise plans, an Owner must enable the connector first. Connect only what you trust, and disconnect what you stop using, because some connectors can also send or share things.
37. Change how Claude Code talks with output styles
What it is: an output style changes how Claude Code responds for the whole session. Built-in styles: Concise leads with the result and skips the recap, Explanatory adds short insight notes, Learning leaves small pieces of code for you to write, and Proactive starts work and makes routine decisions itself. How to switch: in the terminal, type /output-style and pick one; in VS Code, use the slash menu; in the desktop app, set the outputStyle field in your settings file (the value is case sensitive, for example Concise). A custom style is a markdown file in .claude/output-styles. It is an instruction Claude follows, not a guarantee, so use a hook for rules that must always hold. Output styles are a Claude Code feature, not part of claude.ai chat.
38. Talk to Claude with voice mode
What it is: voice mode lets you speak to Claude and hear it answer. How to start: on your phone, tap the sound wave icon next to the text box and pick a voice. On the web or in the desktop app, open a new chat and click the sound wave icon at the bottom right. It listens hands free by default, with a push to talk option. Your conversation is saved as a text transcript, and you can switch between voice and text in the same chat. It is in beta on every plan and counts toward your normal usage limit. Set your spoken language under Settings, then Capabilities. Voice mode is not in Claude Code, where dictation does the job instead.
Plans, Slash Commands, Files and Artifacts: 4 More Tips
39. Do you really need Claude Max? Check your usage first
What it is: Max 5x and Max 20x give you five or twenty times the usage of Pro in each session, and cost about $100 and $200 a month. How limits work: every plan has a session limit that resets every five hours, plus a weekly limit. Anthropic does not publish a fixed message count, and does not promise the weekly limit grows by the same multiple. What to do: before you upgrade, type /usage in Claude Code to see where your usage goes. A long session you never cleared is a common reason. If you hit the wall every day, Max fits. If it happens now and then, usage credits let you keep working past the limit. Check Anthropic’s pricing page for current prices.
40. Save a long prompt as a custom slash command
What it is: a markdown file in the .claude folder of your project, where the file name becomes the command. A file called review.md gives you /review, and everything inside runs. Where to save it: in the project’s .claude/commands folder for that project, or in your home .claude folder to use it everywhere. How to shape it: add a short description at the top, and write $ARGUMENTS where the words you type after the command should go, so /deploy staging fills in “staging”. Newer skill files work the same way, and if a skill and an old command share a name, the skill wins. In the terminal, type the slash and the name. In the desktop app, type a slash in the prompt box, or click the plus button and pick Slash commands.
41. Make Claude create real Excel, PowerPoint, Word and PDF files
What it is: Claude can build the actual file instead of a table you paste by hand. How to switch it on: open Settings, then Capabilities, and under Code execution and file creation turn on Run code in the cloud. It is on by default on most plans, and works on the web, in the desktop app and on your phone. How to use it: just ask, for example “Make an Excel budget with these columns and a total row”, then download the file from the chat. Files can be up to 30 MB, and creating them uses more of your limit than a normal chat, so ask for what you need in one go.
42. Keep your best tools as Claude artifacts
What it is: an artifact is something Claude makes that opens beside your chat, such as a document, a dashboard or a small interactive tool. How to use it: ask for one in any chat, or open the Artifacts tab in the sidebar to see all of yours and start from a template. Change it by asking Claude, and if it shows an error, click Try fixing with Claude. New artifacts start private to you. On paid plans an artifact can remember data between visits, up to 20 MB of text, and it can call Claude itself, so anyone who opens your tool signs in with their own account and uses their own limit, not yours.
Official Docs Power Habits: 10 More Claude Code Tips
These come straight from Anthropic’s Claude Code documentation (best practices, common workflows and cost management). They are the habits that separate a smooth session from a messy one.
43. Give a specific prompt: file, symptom, and what “fixed” looks like
A specific prompt gives Claude information it cannot get from the code. Name the file, describe the symptom, and say what done looks like. “Improve this codebase” triggers broad scanning. “Add input validation to the login function in auth.ts” lets Claude work with minimal file reads, which is both faster and cheaper.
44. Not sure what you want? Ask Claude to interview you
When you have not worked the idea out yet, do not guess a prompt. Say: “I want to add rate limiting to the public API but haven’t thought it through yet. Interview me until you could write the spec, then write it to SPEC.md.” Then build from that spec in a fresh session.
45. Point at files with @ instead of describing them
Type @ and a path to include a file’s contents in your message, for example “Explain the logic in @src/utils/auth.js”. A directory reference shows a file listing, not contents. A file is included when it fits the read limit (25,000 tokens by default), and text files over 256 KB are not included. Referencing a file also brings in the CLAUDE.md files from its folder and parent folders.
46. Resume yesterday’s work instead of re-explaining it
Claude Code saves every conversation locally. Run claude --continue to resume the most recent one in the current folder, claude --resume to choose from a list, or /resume inside a session. For work that spans days, name it with /rename oauth-migration, then later run claude --resume oauth-migration. After opening a pull request, claude --from-pr 1234 (use your own number) finds the session linked to it.
47. Pipe Claude into scripts with -p
Claude can run non-interactively and work like any command-line tool. For example: git log --oneline -20 | claude -p "summarize these recent commits". This is how people use Claude Code in CI, pre-commit hooks and batch jobs.
48. Use a hook to shrink noisy output before Claude reads it
A hook can preprocess data. Instead of Claude reading a 10,000 line log, a hook can search for ERROR and hand back only those lines, cutting tens of thousands of tokens to hundreds. The docs’ example is a PreToolUse hook that rewrites test commands to show only failures. After adding a hook, type /hooks to confirm it is listed.
49. Paste screenshots instead of typing the problem
Drag an image into the window, paste it with Ctrl+V (Alt+V on Windows and WSL), or give Claude a file path. Screenshots of an error, a design mockup or a diagram often explain more than a paragraph of text, and you can add several in one conversation.
50. Ask Claude about Claude, and run /powerup
Claude can answer questions about its own features by looking up the current documentation, for example “how does Claude Code handle permissions?” or “what skills are available?”. Run /powerup for interactive lessons with animated demos.
51. Use a second session as an independent reviewer
One session writes the code, and a second session reviews it without having watched it being written, for example by looking at @src/middleware/rateLimiter.ts. Paste the findings back into the first session. A reviewer that did not write the code is much harder to fool.
52. Habits you can stop
- Shouting in CLAUDE.md (IMPORTANT, ALL CAPS): current models follow a plainly worded line, and heavy emphasis tends to make Claude over-apply it. For a rule that cannot bend, use a hook.
- Typing “think hard” or “be thorough”: that is ordinary text. Effort is the real control, and
ultrathinkworks for one turn. - Describing your architecture in CLAUDE.md: Claude reads the repo itself. Write down only what is not in there.
- Managing the context window by hand all day: most current models have a 1M token window and compact automatically. Running
/clearbetween tasks is still worth it, because a focused context gives better answers and costs less.
Where Each Claude Feature Works
| Feature | Where it works | Good to know |
|---|---|---|
| Projects and memory | Claude chat (claude.ai, desktop, mobile) | Claude Code uses CLAUDE.md instead |
| Research | Web, desktop, mobile (paid plans) | Uses your limit faster |
| Connectors | Web, desktop, mobile | Team and Enterprise: an Owner enables first |
| Claude in Chrome | Google Chrome only (paid plans) | Not mobile, not other browsers |
| Voice mode | iOS, Android, desktop, web | Beta; not in Claude Code |
| Subagents, hooks, plugins | Claude Code: terminal, VS Code, desktop | Some slash commands differ by surface |
| Auto mode | Claude Code: terminal, VS Code, desktop | Not a guarantee; admins can disable on Team and Enterprise |
| Output styles | Claude Code only | Desktop: set in a settings file |
| Cloud sessions, Remote Control | Claude Code (paid plans) | Remote Control needs your computer on |
Claude Code vs Chat and Projects: Which One Should You Use?
| Tool | Best for | Limit to know |
|---|---|---|
| Chat + Projects | Questions, writing, files, shareable artifacts | Cannot act on your computer |
| Claude Code | Real software, version control, parallel work | Needs rules and checks to trust |
Is Plan Mode Still Worth Using?
Yes, because it blocks edits until you approve, which is a safety net. For a big or unclear job, ask Claude to interview you, write a plan to a file, and build it in a fresh session. For a small change you can describe in a paragraph, a short goal, a definition of done and a few guardrails is enough.
How to Start With Claude: A 6 Step Checklist
- Create one Project for the work you repeat most.
- Keep one short CLAUDE.md and add a line after each mistake.
- Pick one model per job and lower effort for easy questions.
- Turn your most repeated prompt into a slash command.
- Add a stop hook for the one rule Claude keeps breaking.
- Start routines on low-stakes jobs only, and watch them for a few weeks before trusting them.
Claude Tips and Tricks FAQ
How do I fix “Claude limit reached”?
Stay on one model per chat, lower the effort level instead of switching models, keep CLAUDE.md under about 200 lines, run /compact early, and check /context to see what fills your window. In Claude Code, a task can also wait for the reset and continue by itself.
Why does Claude forget my instructions?
Each Claude Code session starts fresh, and long chats give Claude more to sift through. Use a Project in chat, keep CLAUDE.md short in Claude Code, and turn must-follow rules into hooks.
What is the difference between Claude skills and MCP?
A skill tells Claude how to do a task. An MCP server or connector gives Claude access to a tool or data source. They solve different problems and often work together.
Are Claude skills useful now?
Yes, if they contain only what Claude cannot work out alone. Test each skill against no skill and remove the ones that do not improve the result.
What is a Claude Code hook?
A hook is a check that runs at a set moment every time, such as when Claude tries to finish. It can be a shell command, an HTTP call, an MCP tool, a model prompt or a subagent.
What is a Claude Code subagent and how do I create one?
A subagent is a helper with its own context window that returns a short summary. Type /agents in Claude Code or ask Claude to create one, and it saves a markdown file with a name, a description, tools and a model in .claude/agents.
What does Claude Auto mode do?
Auto mode lets a safety classifier review each action so you do not approve every step. It blocks risky actions like force pushes and production deploys, and it is not a guarantee.
How do I install a Claude Code plugin?
Type /plugin in Claude Code and open the Discover tab, or run /plugin install with the plugin name and marketplace. In the desktop app, use the plus button next to the prompt box and choose Plugins.
Does Claude have a thinking toggle?
It depends on the model. On the newest models (Haiku 5.5, Sonnet 5.5, Opus 5.5, Fable 5.1 and Opus 5) thinking is locked on, so there is no toggle to find. Use the effort level to control how thorough answers are.
Is Claude Code free?
Claude Code is tied to paid plans and API usage. Check Anthropic’s current pricing page for exact plans, since they change.
Can I use Claude when my laptop is closed?
Yes. Cloud sessions and routines run on Anthropic’s machines, and you can steer them from another device. Remote Control is different: it needs your own computer to stay on.
What is the best Claude tip for beginners?
Create a Project with your reference documents and a short instruction about you. It is the quickest way to stop starting from zero.
Do I need Claude Max?
Check where your usage goes first. Type /usage in Claude Code. If you hit your limit every day, Max fits. If it only happens now and then, usage credits let you keep working past the limit.
How do I resume a previous Claude Code session?
Run claude –continue for the most recent session in the folder, claude –resume to pick from a list, or /resume inside a session. Name a session with /rename and resume it by name.
Can Claude create Excel and PowerPoint files?
Yes. In Settings, under Capabilities, turn on Run code in the cloud for code execution and file creation. Then ask for the file and download it from the chat. Files can be up to 30 MB.
Pick two of these Claude tips and tricks, try them this week, and add a line to CLAUDE.md for anything that still goes wrong.
