Every one of these came off Anthropic's own documentation, not a video. They are grouped by where you type them, so you can skim to the part you actually use. Where a code needs a specific version or plan, it says so on the line.
/help Shows help and the full list of commands available in your setup. Your very first move in a new install, when you want to see what you actually have rather than guess./clear Starts a new conversation with empty context. Pass a name and the old conversation gets that label in the /resume picker. You have finished one job and are starting a completely different one, and you do not want the old thread bleeding in./compact Summarises the conversation so far to free up space. You can pass focus instructions for what the summary should keep. The session is long and slow but you are not done yet, so you want to keep going without losing the thread./resume Reopens an earlier conversation by ID or name, or opens a picker of your past sessions. Alias: /continue. You closed the terminal yesterday and want to pick up exactly where you stopped./model Switches the model and saves it as your default for new sessions. With no argument it opens a picker, and pressing s on a row switches for this session only. You want a cheaper, faster model for routine work and a stronger one for the thinking tasks./effort high Sets how much reasoning the model applies. Accepts low through xhigh, plus max, ultracode and auto. /effort status prints the current level. The answers feel shallow and you would rather wait longer for a better one. max and ultracode last for the session only./usage Shows this session's cost, your plan usage limits and activity stats. /cost and /stats are aliases. You want to know how close you are to your limit before starting something big. On Pro, Max, Team and Enterprise it also breaks down what counts against your plan./config theme=dark Opens the settings panel, or sets a setting straight away when you pass key=value pairs. /config --help lists the keys it accepts. Alias: /settings. You know the one setting you want changed and do not want to hunt through a menu. Other examples in the docs: /config thinking=false and /config model=sonnet./status Opens settings on the Status tab, showing version, model, account and connectivity. Something is behaving oddly and you want to confirm which model and account you are actually on. It works even while Claude is still replying./theme Changes the colour theme, including an auto option that follows your terminal, light and dark variants, and colourblind-accessible themes. You are filming your screen and the default colours do not read well on camera./copy 2 Copies an earlier reply to your clipboard. /copy takes the last one, /copy 2 the second to last. If there are code blocks it shows a picker, and pressing w writes the selection to a file instead. Claude wrote the caption or the paragraph you want and you need it in Notion without mouse-dragging across a terminal./rename Renames the current session and shows the name on the prompt bar. With no name it generates one from the conversation. You have four terminals open and cannot tell which is which./focus Toggles a stripped-back view: your last prompt, a one-line summary of what Claude did, and the final answer. Claude Code only, and only in fullscreen rendering. The wall of tool calls is stressing you out and you only care about the result./plan fix the pricing page Enters plan mode from the prompt. Pass a description and it enters plan mode and starts on that task immediately. You want Claude to think the whole approach through and show you before it touches a single file./permissions Opens a dialog to view, add and remove the allow, ask and deny rules that decide what runs without asking you. You are being asked to approve the same harmless thing twenty times an hour./memory Opens your CLAUDE.md files for editing, toggles auto memory on or off, and shows what auto memory has saved. Claude keeps forgetting a rule you have told it three times, and you want to write it down permanently./add-dir ../brand-assets Adds another folder to the session so Claude can read and edit files in it. Type a partial path and press Tab to accept a suggestion. Your script lives in one folder and the assets live in another, and you do not want to restart./cd ../another-project Moves the session to a different working directory while keeping the conversation. Partial paths autocomplete with Tab. You started in the wrong folder and do not want to lose ten minutes of context restarting./diff Shows every change in your working folder, including the edits Claude has made so far. Before you accept anything, so you can see exactly what changed rather than trusting a summary./doctor Runs a setup checkup that diagnoses problems and offers to fix them: broken installs, unreadable settings files, unused skills and servers, slow hooks, and oversized CLAUDE.md files. It reports first and asks before changing anything. Claude Code is behaving strangely and you have no idea where to start looking./output-style concise Switches the response style. The built-ins are Default, Proactive, Concise, Explanatory and Learning. With no argument it lists them and marks the current one. Needs Claude Code v2.1.269 or later. You want answers that lead with the result and skip the preamble, or the opposite when you are learning something new./skills Lists every skill you have. Type to filter, press t to sort by how much context each one costs, Space or Enter to cycle a skill's visibility, and Esc to save and close. You have installed a pile of skills and want to know which ones are quietly eating your context window./plugin Opens the plugin manager, or takes a subcommand like list, install, enable or disable to act straight away. You want to add a ready-made bundle of skills and connectors instead of building your own./mcp Opens the list of connected servers, or takes reconnect, enable or disable with a server name. Also handles the sign-in flow for servers that need it. Your Notion or Gmail connection has dropped and Claude says the tool is not available./tasks Shows and manages the background work running in this session, including subagents that have finished. Also available as /bashes. You sent Claude off on three things at once and want to see what is actually still running./loop 30m check the deploy Runs a prompt on repeat while the session stays open. Interval and prompt are both optional. Supported units are s, m, h and d. You want something watched on a timer without sitting there refreshing it yourself./loop With nothing after it, runs the built-in maintenance prompt at an interval Claude chooses: finish unfinished work, tend the current branch's pull request, then cleanup passes. Press Esc while it is waiting to stop it. You are stepping away and want Claude to keep tidying up rather than sit idle./schedule Creates, updates, lists or runs routines, which execute in the cloud on a schedule. Claude walks you through setup conversationally. Alias: /routines. You want a report or a check to run every morning whether or not your laptop is open./deep-research what are creators charging for reels in 2026 A bundled workflow that fans out web searches, fetches and cross-checks sources, and writes a cited report. It runs in the background. You need real sources with links, not a confident-sounding summary you cannot verify./dataviz A bundled skill with design guidance for charts and dashboards. It picks the chart form for the data, assigns colour by role, and checks the palette for colourblind safety and contrast. Needs Claude Code v2.1.198 or later. You are making a chart for a carousel or a media kit and want it to look designed rather than default./design a pricing page for a 30 day cohort A bundled skill that drafts mockups, screen flows, landing pages or posters as artboards on one canvas, published as a Design artifact you edit in a browser and export as PNG or PDF. Needs Claude Code v2.1.265 or later and a session where artifacts are available. You want a visual draft to react to instead of describing a layout in words./artifacts Lists the artifacts you own or that are shared with you, then lets you attach one to the session, open it in your browser, or copy its link. Needs Claude Code v2.1.208 or later. You made a page last week, want to send the link, and cannot remember where it went./insights Generates an HTML report on your recent sessions on this machine: which projects you work in, how you use Claude Code, where things go wrong, and features to try. Not available in cloud sessions. You want an honest read on your own habits before deciding what to automate next./team-onboarding Reads your last 30 days of sessions, commands and server usage and writes a markdown onboarding guide a teammate can paste as their first message. Pro, Max, Team and Enterprise claude.ai subscribers also get a share link. You are handing a workflow to an editor or a VA and do not want to write the setup doc yourself./powerup Walks you through Claude Code features as short interactive lessons with animated demos. You suspect you are using about ten percent of the tool and want the guided tour rather than the docs./statusline Configures the status line under your prompt. Describe what you want, or run it with no arguments and it auto-configures from your shell prompt. You want the model, the folder and your usage visible at a glance while you film.Shift+Tab Cycles permission modes: default (shown as Manual) to acceptEdits to plan and back. Optional modes slot in after plan when enabled. The fastest way to flip into plan mode before a big change, or into accept-edits once you trust the direction.Ctrl+C Interrupts whatever is running. If nothing is running, the first press clears your prompt and a second press exits Claude Code. Claude has gone off in the wrong direction and you want it to stop right now.Ctrl+L Forces a full terminal redraw, keeping your input and conversation history. The screen has gone garbled or half-blank and you want it clean again without losing anything.Ctrl+O Toggles the transcript viewer, showing detailed tool usage with a timestamp and the model used on each reply, and expanding lines that collapse by default. You want receipts on what Claude actually did, not the tidy summary.Ctrl+R Reverse-searches your prompt history. Press it again to cycle older matches, Tab or Esc to accept and keep editing, Enter to run it. You wrote the perfect prompt last Tuesday and want it back without retyping it.Ctrl+B Moves a running Bash command or agent to the background so you can keep working. Tmux users press it twice. Something is taking four minutes and you do not want to sit and watch it.Ctrl+T Shows or hides Claude's to-do checklist in the status area. You want to see the plan it is working through without asking it to repeat itself.Ctrl+G Opens your prompt in your default text editor. Ctrl+X Ctrl+E is the same thing in readline form. You are writing a long, structured prompt and the single-line terminal box is fighting you.Ctrl+V Pastes an image from your clipboard and drops an [Image #N] chip at the cursor so you can refer to it by position. Use Cmd+V in iTerm2, or Alt+V on Windows and WSL. You screenshotted a reference and want Claude to look at it rather than read your description of it.Ctrl+Enter Sends your queued messages right away instead of waiting for the current turn to end. Ctrl+X Ctrl+S does the same in any terminal. You typed a correction while Claude was working and want it to land now, before more wrong work piles up.Shift+Enter Adds a new line instead of sending. Native in iTerm2, WezTerm, Ghostty, Kitty, Warp, Apple Terminal and Windows Terminal. You are writing a multi-paragraph brief and keep sending it by accident halfway through.\ then Enter The universal way to add a new line to your prompt. Works in every terminal with no setup. Shift+Enter does not work in your terminal and you do not want to configure anything.Ctrl+J Inserts a new line using a control sequence. Works in any terminal without configuration. The backslash trick annoys you and Shift+Enter is not supported where you are.Option+P Switches model without clearing what you have already typed. Alt+P on Windows and Linux. You are halfway through a prompt and realise it needs the bigger model.Option+T Toggles extended thinking. Alt+T on Windows and Linux. No effect on Opus 5.5 or the Fable models, which always use extended thinking. The task needs care and you want the model to slow down for it.Option+O Toggles fast mode on or off. Alt+O on Windows and Linux. You are doing quick repetitive work and want speed over depth.Tab Accepts the highlighted autocomplete suggestion, including the greyed-out next-prompt suggestion. Right arrow also accepts a suggestion. Claude has guessed your next step correctly and you would rather press one key than type the sentence.Up arrow Recalls earlier prompts, reaching back into past sessions in the same folder. From the first line of the input box it also pulls back messages you have queued so you can edit them. You want to rerun a prompt from a previous session, or take back something you queued by mistake.Esc Stops the current response or tool call mid-turn so you can redirect. The work done so far is kept. Claude is confidently building the wrong thing and you want to cut in without killing the session.! npm test A prompt starting with ! runs the command directly without Claude interpreting or approving it, adds the output to the conversation, and Claude then responds to that output automatically. You want to run something yourself and have Claude explain the result, in one move instead of two.@ Triggers file path autocomplete so you can point Claude at a specific file. Resources from connected servers appear in the same list. You want Claude reading the one file you mean, not searching the whole folder for it.@github:issue://123 References a resource from a connected server using @server:protocol://resource/path. It is fetched and attached automatically, and you can reference several in one prompt. You want Claude to pull a specific record from a connected tool rather than describing it yourself.@agent-code-reviewer Guarantees a specific subagent runs for one task. Type @ and pick from the typeahead, or type @agent-<name> manually. Plugin subagents use their scoped name. You have set up a specialist and want it used for this task rather than leaving the choice to Claude.:heart: Typing a full emoji shortcode inserts the emoji as soon as you type the closing colon. Two or more characters after the colon opens a suggestion popup. Needs Claude Code v2.1.217 or later. You are drafting a caption in the terminal and want the emoji without leaving the keyboard.? Typed on an empty input, toggles the keyboard shortcut help panel. In the transcript viewer it shows the shortcuts available there. You have forgotten a shortcut and do not want to open a browser tab to find it.run the tests, then /com Command completion works partway through a prompt: type / after a space, then the first letters of a command name, and press Tab to insert the top match. You want to chain a command onto the end of an instruction instead of sending two separate messages.claude Starts an interactive session in the current folder. The plain start. Everything else on this list is a variation on it.claude "explain this project" Starts an interactive session with that first prompt already sent. You already know your opening question and do not want to type it twice.claude -c Loads the most recent conversation in the current folder. You closed the terminal by accident and want the last thread back in one keystroke.claude -r "auth-refactor" Resumes a specific session by ID or name, or opens a picker when you give it nothing. You have several named sessions on the go and want a particular one.claude -p "summarise this file" Runs one prompt, prints the answer, and exits without an interactive session. You want a one-shot answer you can pipe into another command or a script.claude --model sonnet Sets the model for this session only, using an alias like sonnet, opus, haiku or fable, or a full model name. You want a specific model for one job without changing your saved default.claude --permission-mode plan Starts the session already in plan mode. The flag also accepts default, acceptEdits, auto, dontAsk and bypassPermissions. You want the first thing that happens to be a plan, not an edit.claude --safe-mode Starts with every customisation disabled: CLAUDE.md, skills, plugins, hooks, servers, custom commands, agents, output styles, themes, keybindings and status line. Something is broken and you need to know whether it is Claude Code or something you added.claude --chrome Turns on the Chrome browser integration for this session so Claude can navigate, click and type in your browser. You want Claude to actually go and check a page rather than guess what is on it. Run /chrome to make it the default.claude agents Opens agent view, where you monitor and dispatch parallel background sessions. Press ? inside it to see every shortcut. You have work running in the background and want one screen that shows all of it.claude update Updates Claude Code to the latest version. A feature you read about is missing and you are probably a few versions behind.claude doctor Prints read-only installation and settings diagnostics from the terminal without starting a session. Claude Code will not start properly and you need the diagnosis before you can open a session to ask for one.claude mcp add --transport http notion https://mcp.notion.com/mcp Connects a remote server over HTTP, the recommended transport. Add --header to pass a token. You want Claude reading and writing in Notion, and this is the real command, copied from the docs.claude mcp list Lists every configured server with a health status next to each one, such as connected or failed to connect. claude mcp get <name> shows the detail for one. A connection is not working and you want to see which one, before you start guessing.~/.claude/CLAUDE.md Your personal instructions file. It loads in every project on your machine, at the start of every session. Anything true about how you work everywhere: your tone rules, your banned words, your naming conventions../CLAUDE.md Project instructions that load in every session in that folder. Can also live at ./.claude/CLAUDE.md. Aim for under 200 lines, because longer files eat context and get followed less closely. The facts about one project that you keep re-explaining. Run /init to generate a starting version.CLAUDE.local.md Personal instructions for one project, loaded alongside CLAUDE.md. Add it to .gitignore so it is never shared. Your own preferences on a shared project, the ones nobody else needs to see.@docs/style-guide.md Inside a CLAUDE.md, an @path line pulls that file in and loads it at launch. Relative and absolute paths both work, and imports can nest up to a depth limit. Wrap a path in backticks to mention it without importing it. Your instructions file has grown into a monster and you want it split into readable pieces..claude/rules/testing.md A folder of separate instruction files, one topic per file. Every .md file in it is discovered, and files without a paths field load every session at the same priority as .claude/CLAUDE.md. One instructions file is no longer enough and you want tone, structure and process kept apart.paths:
- "src/api/**/*.ts" YAML frontmatter at the top of a rules file that scopes it to matching files, so the rule only applies when Claude is working on them. You have a rule that only matters for one part of the project and you do not want it loaded the rest of the time.{ "autoMemoryEnabled": false } Put this in ~/.claude/settings.json to stop Claude writing its own notes about your corrections and preferences. The /memory toggle writes the same key. You would rather control exactly what Claude remembers, in files you wrote yourself.~/.claude/loop.md Replaces the built-in maintenance prompt that a bare /loop runs. .claude/loop.md in a project takes precedence. Edits take effect on the next iteration. You want your own definition of what Claude should do with spare cycles.description: Summarises uncommitted changes and flags anything risky. Use when the user asks what changed. The one frontmatter field worth writing. Claude uses it to decide when to apply the skill. If you leave it out, the first non-empty line of the file is used instead. Every skill you make. A vague description is the single most common reason a skill never fires.disable-model-invocation: true Stops Claude loading the skill on its own, so it only runs when you type /name. A workflow you want to fire deliberately, not whenever Claude thinks it is relevant.allowed-tools: Read Grep Lists tools the skill may use without asking you for permission. The grant lasts for that turn only and clears when you send your next message. A skill that always needs the same two permissions and you are tired of approving them.$ARGUMENTS Inside a skill, gets replaced by whatever you typed after the skill name. If you pass arguments and there is no placeholder, the text is appended as ARGUMENTS: <your input> so Claude still sees it. Turning a fixed prompt into a reusable one: /fix-issue 123 instead of a skill that only ever handles one thing.$0 $1 $2 The short form of $ARGUMENTS[0], $ARGUMENTS[1] and so on, for reading arguments by position. A skill that takes several inputs in a set order, like a file and a format.!`git diff HEAD` A line like this inside a skill runs the command and replaces the line with its output before Claude reads the file, so the instructions arrive already grounded in live data. A skill that should always work from what is true right now, not from what you described when you wrote it./plugin marketplace add anthropics/claude-plugins-official Registers a plugin catalogue with Claude Code so you can browse it. Nothing is installed by this step. The official marketplace is missing, usually because your network blocked it on first launch./plugin install github@claude-plugins-official Installs one plugin from a marketplace. Browse what is available with /plugin and the Discover tab. You want a ready-made bundle of skills, agents and connectors instead of wiring each piece up yourself.Double-tap Option Opens Claude quick entry from any app on a Mac, as a text box you type into. Claude Desktop has to be running, but it can be in the background. An idea lands while you are editing a video and you want it captured without switching apps. Claude Desktop on macOS only.Option + Space The alternative quick entry shortcut, set in Settings > General under Desktop app, next to Quick access shortcut. You can also set a fully custom shortcut there. Double-tapping Option clashes with something else you use.Caps Lock Starts voice dictation in quick entry with real-time transcription, and stops it when you press it again. Off by default, because it takes over your Caps Lock key. Needs macOS 14 or later. You think out loud better than you type, and you want the idea down before it goes./ Typed in the chat box on claude.ai, opens the same menu as the + button in the lower left, where connectors live. You want to turn a connected tool on for this conversation without reaching for the mouse.remember this Tells Claude to save something to memory directly, rather than waiting for it to decide on its own. Memory is on by default on Free, Pro and Max; on Team and Enterprise an owner turns it on. A fact you will need in every future chat, like a deadline moving or how you want updates written.claude://claude.ai/new?q=Summarize%20this%20week Opens a new chat in Claude Desktop with the prompt field prefilled. All parameter values must be URL-encoded, and q is truncated at roughly 14,000 characters. Building a one-click button in your own tools, a shortcut, or a bookmark that starts the same chat every time.claude://code/new?q=Fix%20the%20test&repo=owner%2Fname&branch=main Opens the new Claude Code session composer on mobile, optionally prefilled. mode accepts plan or code, repo takes owner/name, and branch needs repo. Requires Claude Code access on your account. Kicking off a job from your phone, from a shortcut or a link, without typing it all out on a small keyboard.Ask Claude An iOS intent you can trigger from Spotlight search, Siri, the Share menu, or any app that supports intents, and chain inside the Shortcuts app. Needs iOS 18 or above. Sending selected text straight to Claude from whatever app you are in, or building a one-tap summarise shortcut.Claude does not keep a hidden list of secret commands. When you type /godmode, it reads that the way it reads any other words in your message and makes its best guess at what you meant. So a code works exactly as well as the word is clear, and stops working the moment the word is doing the heavy lifting on its own.
Anthropic's own prompting guidance is unusually blunt about this. It says Claude "responds well to clear, explicit instructions", that being specific about your desired output helps, and that if you want above and beyond behaviour you should "explicitly request it rather than relying on the model to infer this from vague prompts". The golden rule it gives is the one worth remembering: show your prompt to a colleague with minimal context, and if they would be confused, Claude will be too.
ultrathink anywhere in your prompt requests deeper reasoning on that turn, and Claude Code recognises it and adds an in-context instruction. Phrases like "think", "think hard" and "think more" are passed through as ordinary prompt text and are not recognised as keywords. That is documented on the model configuration page.Every one of these replaces a code people type. Paste the prompt, then your actual question or your actual text underneath it. They work in any Claude, on any plan.
1. Instead of /godmode, meet the decisive advisor. The code was promising a blunt, unhedged answer. Here is the version that reliably produces one.
Answer this as a decisive expert advisor who has to live with the consequences, not as an assistant listing options.
Lead with one clear recommendation in your first sentence, then give the three strongest reasons for it, most important first. Name the single biggest risk in that recommendation and one concrete way to reduce it, then tell me what you would cut or stop doing. Do not hedge: if the honest answer is that it depends, say exactly what it depends on and then still pick the option you would choose. If my plan is the wrong call, say so in the first sentence.
Here is the decision:2. Instead of /10x, meet the sharpener. Ten times what, though? The code never says. Naming the exact upgrade is the whole fix.
Rewrite what follows so it is sharper, not longer. Give me five versions, each taking a genuinely different angle rather than five rewordings of the same sentence.
Hold every version to these constraints: shorter than the original, one concrete noun or number in it, no adjectives doing work a noun could do, and no phrase that could appear in anybody else's version of this. Under each one, write a five-word note on the specific mechanism it is using, for example a number, a contradiction, a named person, a consequence or a question. Then tell me which one you would actually publish and why.
Here is the text:3. Instead of /ELI5, meet the plain-English teacher. This is the one that already half worked, because it names an audience and a reading level. This version pins it down properly.
Explain the following to me as if I am smart but completely new to this field, using no jargon at all in the first paragraph.
Start with one everyday analogy that makes the shape of the idea obvious, then explain how the real thing works in plain words, then name the two or three technical terms I will meet if I read more about this and define each one in a single sentence. Finish with the one misunderstanding beginners usually have about it and why it is wrong. Do not flatter me, do not pad, and stop when the idea has landed.
Here is the thing to explain:4. Instead of /ooda, meet the fast decider. OODA is a real decision loop, observe, orient, decide, act, developed by the military strategist John Boyd. The code assumes Claude will run it properly. This prompt makes it.
Run the OODA loop on the situation below and label each of the four stages out loud as you go.
Observe: list only the facts I have actually given you, and separately list the facts you would need but do not have. Orient: say what these facts most likely mean, name the assumption you are least confident about, and state one plausible alternative reading of the same facts. Decide: pick one course of action and say why it beats the next best option. Act: give me the first three steps, the first one being something I can do in the next hour, and the signal that would tell me to abandon this course and re-run the loop.
Here is the situation:5. Instead of /ghost, meet the de-robotiser. The code promised writing that beats AI detectors. What it can honestly do is take the default-AI sound out of your writing.
Rewrite the text below so it reads like a real person wrote it, keeping every fact, number and claim exactly as they are. Change only the voice.
Remove the giveaways: em dashes, the "it's not X, it's Y" construction, everything bundled into tidy groups of three, and words like delve, tapestry, landscape, pivotal and testament. Vary the sentence lengths deliberately, let one or two sentences run long and one be very short, and leave in the small redundancies a person would actually say out loud. Do not make it more formal. When you are done, list the three specific changes that did the most work, so I can make them myself next time.
Here is the text:6. Instead of /L99, meet the senior expert. The promise was an expert-level answer. Claude does not have a level dial, but naming the expert and the standard gets you there.
Answer this the way a genuinely senior specialist in this exact field would answer a peer, not the way a general explainer would answer a beginner.
Use the real technical vocabulary of the field and define a term only the first time it appears. Tell me what the current consensus is and where practitioners actually disagree. Name the specific thing that separates a good answer here from an obvious one, and point out anything in my question that shows I am framing the problem in an unhelpful way. If the honest answer is that nobody knows, say that rather than producing something that sounds authoritative.
Here is the question:7. Instead of /artifacts, meet the builder. Artifacts are a real Claude feature, not a code. You get one by describing something worth building, or by choosing Output in the message box. Here is the ask that reliably produces a working one.
Build this as a single self-contained artifact I can use straight away, not a description of how I could build it.
Make it work end to end with realistic sample data already filled in, so I can see what it does the moment it loads. Use a clean, modern layout that works on a phone screen as well as a laptop, and make every number and label editable by me rather than hard-coded into the logic. When you are finished, tell me in three lines what it does, the one thing it deliberately does not handle, and the single most useful change I could ask you for next.
Here is what I want:More on how artifacts work, and the Output button that creates one: Claude's own artifacts guide.
8. Instead of /devil, meet the devil's advocate. The code asks for pushback. This version makes the pushback specific enough to be useful.
Argue against the plan below as hard as you honestly can. Do not soften it, and do not open by agreeing with me first.
Give me the three strongest objections, ranked by how likely they are to actually sink this, and for each one name the specific assumption it attacks and what evidence would prove it right or wrong. Then tell me the one objection you think I am most likely to dismiss too quickly, and why. Finish with the smallest, cheapest test I could run this week that would settle the biggest disagreement between us.
Here is the plan:This is the part that makes the whole idea real. In Claude Code, a skill's directory name becomes the command you type, so a folder called godmode with a SKILL.md inside it gives you an actual /godmode command that shows up in the suggestion list. No plugin, no paid plan, no code.
~/.claude/skills/godmode/SKILL.md. The file has to be called SKILL.md, and the folder has to be called whatever you want to type. A personal skill saved there works across all your projects on that machine./, and /godmode appears in the suggestion list. Claude Code watches the skills folder for changes, so you do not need to restart to pick up a new skill.godmode, make sure the file itself is named SKILL.md, and zip the folder so the skill folder is the root of the zip rather than sitting inside a subfolder. Then open Claude, go to Customize, then Skills, click Add, choose Upload a skill and upload the zip./blunt or /sharpen instead? Change the folder name and that is the new command. Drop any of the eight prompts above into the body of the file and you have that persona as a permanent command.---
description: Direct, decisive advisor mode. Use when the user types /godmode, or asks for a blunt recommendation, a straight answer, or a decision instead of a list of options.
---
# God mode
The user wants a call, not a menu. Answer as a decisive expert advisor who has to live with the consequences of the recommendation.
## Instructions
1. Lead with one clear recommendation in the first sentence. No preamble, no restating the question back.
2. Give the three strongest reasons for it, most important first, one line each.
3. Name the single biggest risk in this recommendation, and one concrete way to reduce it.
4. Say what you would cut, skip, or stop doing. A real decision always leaves something behind.
5. Do not hedge. If the honest answer is "it depends", say exactly what it depends on, then still pick the option you would choose.
6. If something you genuinely need is missing, ask one question. Then give your best call anyway, with your assumption stated in one line.
7. If the user's stated plan is the wrong call, say so in the first sentence and give the alternative. Being agreeable is not the job.
## Format
Keep it under 200 words unless the user asks for more. No headings. No list longer than three items. No closing summary and no "let me know if you would like me to expand on any of this".~/.claude/skills/synced/ for skills downloaded from claude.ai and will skip a skill you author there.
The actual list of Claude Code slash commands
Anthropic on clear, explicit instructions and the golden rule
ultrathink is a keyword, "think hard" is not
Skills in Claude Code, including how a skill gets its command name
Creating and uploading a custom skill in the Claude app
What artifacts are and how to get one
The OODA loop, the real decision framework behind that code
the waitlist is now open
A 30-day cohort to build real AI systems with me, step by step. Join the waitlist and you hear first, and lock the founding price.