You already know how to work with an AI agent in the terminal. This guide simply maps what you know from Claude Code to what you'll find in GitHub Copilot CLI, so you're productive on day one. Every command below is read straight from the CLI itself, so you can trust it.
Disclaimer: This is a personal/community guide, not official GitHub documentation. For the canonical source see docs.github.com/copilot. Command descriptions are taken verbatim from the CLI's own command list (v1.0.59).
- Same kind of tool. A terminal-native AI agent that reads and edits your files, runs shell commands, and works in a loop until the job is done.
- Same brains, your choice.
/modellets you pick Claude Sonnet/Opus, GPT-5, Gemini, and more, all in one place. - Deeper GitHub power. Hand a whole task to the cloud and get a pull request back (
/delegate), run many agents at once (/fleet), steer it from your phone (/remote), and do real research with citations (/research). - It schedules work for you.
/everyand/afterlet your agent run tasks on a timer, like an alarm clock for your code. - It checks its own work.
/rubber-duckgets a second opinion,/security-reviewhunts for vulnerabilities, and/reviewdoes code review.
In short: same brain if you want it, a much bigger toolbox, and tighter GitHub integration.
Before the command lists, here's why this is worth your time. Imagine you don't have one assistant, but a small team of agents, each with a job:
- An operations agent. Every Monday it drafts your team's weekly plan on its own, pulling the latest status updates, priorities, and ongoing workstreams from your
team-opsnotes. Nobody has to compile all of that by hand every week. - A culture agent. It reads a
team-culture-playbook, so the update isn't dry and robotic. It knows how your team likes to communicate: the language you use, how decisions are framed, what behaviours the team values. Process alone isn't enough; teams run on culture and shared ways of working. - A tone-and-personality agent. It follows something like a
rob-leadership-styleguide so the message still sounds like your team. Some leaders are direct, some are highly empathetic, some are very structured. The agent adapts its tone so the output feels familiar and trusted, not like it came from a machine.
Why bother? People spend a huge amount of time on the same coordination work over and over: gathering context, rewriting the same updates, aligning messaging, keeping things consistent across teams. These agents quietly absorb that overhead, so instead of you doing the repetitive work, they continuously gather context, structure it, and communicate the way your team already operates.
The exciting part: not long ago this was just a "wouldn't it be nice…" idea. With Copilot CLI you can actually start building agents like this yourself, where you define their instructions, prompts, schedules, and the context they read from. And as long as the machine is running (a DevBox or any always-on computer), they keep working on their own.
And it's all built from the same small pieces you're about to learn: an agent's "personality" is just an instructions file (§5), "every Monday" is scheduling with /every and /after (§2), and "a team of agents" is /agent + /fleet. You don't need all of this on day one, but this is where it's heading.
And here's how little it takes. You're not writing code; you're writing plain English. Roughly:
# 1. Give each agent its "personality", just a plain-text file of instructions.
# Drop these context files in your project (they're read automatically, see §5):
team-ops.md ← what to pull: status, priorities, workstreams
team-culture-playbook.md ← how your team talks and makes decisions
rob-leadership-style.md ← the tone to write in
# 2. Turn it into a real, reusable agent: one file in .github/agents/
# Filename is the agent's name, e.g. .github/agents/weekly-update.agent.md
---
name: weekly-update
description: Drafts our weekly team update in our own voice.
tools: ["*"] # which tools it may use (* = all). Optional.
# model: claude-sonnet-4.5 # optional: pick a brain (any id from /model)
---
You are our weekly-update writer. Use team-ops.md for what happened,
team-culture-playbook.md for how we phrase things, and rob-leadership-style.md
for tone. Keep it warm and to the point, and save the result to updates/.
# 3. Start Copilot in that folder, then pick your agent
copilot
/agent ← lists your agents; choose "weekly-update"
# 4. Make it run on its own, every Monday-ish
/every 1d if it's Monday, run the weekly-update agent
# 5. (optional) Run several agents at once
/fleet
That's the whole trick: a few context files, one small agent file, and a /every timer. If you can write a message to a teammate, you can set this up.
You do not need to memorize the big table below. For vibe-coding, these 10 moves are enough. Everything else is bonus power you can grow into:
| # | Do this | What it's for |
|---|---|---|
| 1 | copilot |
Start it in any folder |
| 2 | /model |
Pick your AI brain (Claude, GPT-5, Gemini…) |
| 3 | @file |
Show it a file |
| 4 | !command |
Run a terminal command |
| 5 | /init |
Teach it your project's "house rules" |
| 6 | /diff |
See what changed |
| 7 | /undo (or /rewind) |
Take back the last step |
| 8 | /compact |
Shrink a long chat to stay fast |
| 9 | /resume |
Jump back into an earlier chat |
| 10 | /help |
See everything |
📋 A quick note on availability. Most commands work for everyone. A few newer ones are still rolling out, so they may not be switched on for every account yet. If you type a command tagged
⚠️ and see "Unknown command," don't worry, it's not a mistake on your end, it just isn't enabled for you yet. Everything untagged works for everyone.
If your fingers already know Claude Code, here's the quick translation. Most things are the same; a few have new names.
| You used to do this in Claude Code | In Copilot CLI you do this |
|---|---|
claude to start |
copilot to start |
/login |
/login ✅ same |
/model to switch models |
/model ✅ same |
/clear to reset chat |
/clear (abandons the session) or /new (fresh chat, keeps the session) |
/compact to shrink context |
/compact ✅ same |
/resume a previous session |
/resume ✅ same |
/init to create project memory |
/init ✅ same (creates Copilot instructions) |
CLAUDE.md for project memory |
CLAUDE.md, AGENTS.md, or .github/copilot-instructions.md, all are honored |
/agents to manage subagents |
/agent (browse) + /fleet (parallel) + /tasks (manage) |
/mcp for MCP servers |
/mcp ✅ same |
/permissions |
/allow-all, /add-dir, /list-dirs, /reset-allowed-tools |
/cost / usage |
/usage |
/help |
/help ✅ same |
/exit |
/exit ✅ same |
/chrome (browser control) |
❌ Not built-in, add it through /mcp (e.g. chrome-devtools-mcp or playwright-mcp) |
/ide to connect your editor |
/ide ✅ same |
/review for code review |
/review ✅ same |
/pr workflow |
/pr ✅ same, more GitHub-native |
@file to attach a file |
@ to mention files ✅ same |
#issue mentions |
# to mention issues & PRs ✅ same |
!command shell escape |
! to run a shell command ✅ same |
shift+tab to switch modes |
shift+tab ✅ same |
Migration tip: your existing CLAUDE.md works as-is. No rewrite needed.
How to read this table: The middle column is the exact words Copilot shows you when you type
/. The right column is the same idea in plain English, with an example of how to type it. Some commands take extra input (shown in< >), just type the command, a space, then your text.
👤 Designers / vibe-coders, relax: anything tagged 🔧 is engineer-focused setup or plumbing. It's nice to know it exists, but you don't need it to vibe-code, so feel free to skip those rows. Everything else is useful to know, though your exact
/helplist can vary by account, rollout, and organization policy.
Legend (what the icons mean):
| Icon | Meaning |
|---|---|
| ⭐ | Star command: an especially useful, "don't miss this" feature worth trying first. |
| 🔧 | Engineer / setup: plumbing for developers. Designers can safely skip these. |
| Newer / rolling out: works on most accounts, but may show "Unknown command" if not yet enabled for yours. Not a bug. |
Tip on arguments: most commands you just type on their own. Some accept extra input after the slash, type the command, a space, then your text or one of the options. Each command below spells out its options. (Intervals/delays look like
30s,5m,2h,1d.)
| Command | What Copilot shows you | In plain words (+ how to type it) |
|---|---|---|
/init |
Initialize Copilot instructions for this repository | Creates a "house rules" file so Copilot remembers how your project works. Type /init once per project. Add /init suppress to do it quietly (it still writes the file, just without the chatty output). |
/agent |
Browse and select from available agents (if any) | Picks a specialist helper for the job, like choosing the right expert from a list. Type /agent and choose one. To create one, add a file like .github/agents/my-agent.agent.md with a short name: / description: header and your instructions in the body, then it shows up here. |
/skills |
Manage skills for enhanced capabilities | Turns extra abilities on or off (like add-on powers). Options: list (see all skills)info (details of one)reload (refresh after changes). |
/mcp |
Manage MCP server configuration | Plugs outside tools into Copilot, like a browser, a database, or Figma. Think of it as adding new apps to your phone. Options: list (see connected tools)show (details of one)enable / disable (turn a tool on/off)reload (refresh). |
/plugin |
Manage plugins and plugin marketplaces | A plugin is a bundle of extra features you install from a store (a "marketplace"). Options: list (see installed plugins)marketplace add / remove / list / browse (manage where plugins come from). |
| Command | What Copilot shows you | In plain words (+ how to type it) |
|---|---|---|
/model |
Select AI model to use | Choose which AI brain runs: Claude, GPT-5, Gemini, etc. Auto lets Copilot pick for you, and one model is marked (default). Type /model and pick one. ⭐ |
/delegate |
Send this session to GitHub and Copilot will create a PR | Hand the whole job to GitHub's cloud: it does the work on its own computers and hands you back finished changes (a "pull request") to review. You don't have to babysit it. Needs a GitHub repo. Type /delegate. ⭐ |
/fleet |
Enable fleet mode for parallel subagent execution | Puts several helpers to work at the same time instead of one, so big jobs finish faster, like a team splitting up chores. Type /fleet then describe the work. ⭐ |
/autopilot |
Toggle autopilot mode or set an explicit objective | Lets Copilot keep working on its own until the goal is met. Options: on / off (turn it on or off)or type a goal like /autopilot fix all failing tests (giving an explicit goal may be rolling out). ⭐ |
/tasks |
View and manage tasks (subagents and shell commands) | A dashboard of everything running right now (background helpers and commands), so you can check on or stop them. Type /tasks. |
/rubber-duck |
Get an independent critique of your current work from the rubber duck agent | Asks a second, independent AI to double-check the plan or work and catch mistakes you (or the first AI) might miss. Named after "rubber duck debugging." Type /rubber-duck (optionally add a question to focus on). ⭐ |
| Command | What Copilot shows you | In plain words (+ how to type it) |
|---|---|---|
/ide |
Connect to an IDE workspace | Links Copilot to your code editor (VS Code) so they share the same files and you see changes live. Type /ide. |
/diff |
Review the changes made in the current directory | Shows exactly what was changed, line by line (added lines and removed lines), so you can see what the agent did before keeping it. Type /diff. |
/pr 🔧 |
Operate on pull requests for the current branch | A pull request (PR) is how you propose your changes to be added to a project. This creates, views, or works on one. Type /pr (works when you're in a GitHub project). |
/review |
Run code review agent to analyze changes | A robot reviewer reads through your changes and points out real problems (bugs, mistakes) before anyone else sees them. Type /review (optionally add what to focus on). ⭐ |
/security-review 🔧 |
Analyze staged and unstaged changes for security vulnerabilities. | Scans your changes for security holes before you ship. Type /security-review (optionally add instructions). ⭐ |
/lsp 🔧 |
Manage language server configuration | A language server is a live code expert for one language (like Python or TypeScript), and it lets Copilot jump to definitions, spot errors, and edit accurately, the way a code editor does. This manages those. Type /lsp. |
/plan |
Create an implementation plan before coding | Asks Copilot to make a step-by-step plan first, so you can check the approach before it starts changing anything. Type /plan then what you want, e.g. /plan build a login page. |
/terminal-setup 🔧 |
Configure terminal for multiline input support (shift+enter) | One-time setup so shift+enter makes a new line. Type /terminal-setup once. |
| Command | What Copilot shows you | In plain words (+ how to type it) |
|---|---|---|
/allow-all |
Enable all permissions (tools, paths, and URLs) | Stops asking "is this OK?" for everything: fast, but only use it in folders you trust (it can run any command). Type /allow-all. |
/add-dir 🔧 |
Add a directory to the allowed list for file access | Lets Copilot touch one extra folder. Type /add-dir ./my-folder. |
/list-dirs 🔧 |
Display all allowed directories for file access | Shows which folders Copilot is allowed to use. Type /list-dirs. |
/cwd 🔧 |
Change working directory or show current directory | Shows or changes the folder you're working in. Type /cwd to see it, or /cwd ./project to move. |
/reset-allowed-tools 🔧 |
Reset the list of allowed tools | Forgets all the "yes, allow" answers and starts fresh. Type /reset-allowed-tools. |
/sandbox 🔧 |
Configure sandbox modes | Puts a safety wall around the agent: it can work inside your project folder, but can't touch the rest of your computer. Like saying "you can play in this room, but the other doors are locked." Options: enable (wall on)disable (wall off). |
| Command | What Copilot shows you | In plain words (+ how to type it) |
|---|---|---|
/resume |
Switch to a different session (optionally specify session ID, task ID, or name) | Re-opens an earlier chat so you can pick up where you left off. Type /resume and choose from the list, or add a name/ID. |
/rename |
Rename the current session | Gives this chat a memorable name. Type /rename demo-prep. |
/context |
Show context window token usage and visualization | Shows how full the AI's short-term memory is right now, like a battery bar for how much it can still keep in mind. Type /context. |
/usage |
Display session usage metrics and statistics | Shows how much you've used in this chat (requests, etc.), like checking your data usage. Type /usage. |
/session |
View and manage sessions. Use subcommands for details. | Inspect or organize your chats. Options: info (this chat's details)checkpoints (saved restore points)files (files this chat touched)plan (the current plan)rename (give it a name). |
/compact |
Summarize conversation history to reduce context window usage. Optionally provide focus instructions. | When a chat gets long and slow, this shrinks it into a short summary so the AI stays fast, like zipping a big file. Type /compact, or /compact keep the login details. |
/share |
Share session or research report to markdown file, HTML file, or GitHub gist | Saves this chat so you can send it to someone, as a file or a shareable web link. Choose what: a session (this chat) or a research report. Choose the format: a Markdown file, an HTML file, or a GitHub gist (a quick shareable link). |
/remote |
Show remote status or toggle remote control from GitHub web and mobile | Lets you steer this session from your phone or the web. Options: on (allow phone control)off (stop it)show (check status). ⭐ |
/copy |
Copy the last response to the clipboard | Copies the last answer so you can paste it. Type /copy. |
/rewind |
Rewind the last turn and revert file changes | Undo button: takes back the last step and its file edits. Type /rewind. |
/undo |
Rewind the last turn and revert file changes | Same as /rewind. Type /undo. |
| Command | What Copilot shows you | In plain words (+ how to type it) |
|---|---|---|
/every |
Schedule a recurring prompt or skill for this session | A repeating timer for your agent. Interval looks like 30s, 5m, 2h, 1d. Example: /every 10m run the tests and tell me if anything broke. ⭐ |
/after |
Schedule a one-shot prompt or skill for this session | A one-time delayed task. Example: /after 30m remind me to push my code. ⭐ |
/memory |
Show memory status, or enable/disable memory across sessions | Lets Copilot remember useful facts about you and your work between chats, so you don't repeat yourself. (It remembers facts, not whole conversations.) Options: on (start remembering)off (stop)show (see what it knows). |
/subconscious 🔧 |
Manage Copilot Subconscious memory consolidation | Tidies up what Copilot remembers in the background, like your brain sorting memories while you sleep. Args: run. |
/keep-alive |
Manage keep-alive mode (prevents system sleep). | Stops your Mac from going to sleep during a long job, so the agent isn't interrupted when you step away. (Doesn't help if you close the lid.) Type /keep-alive to toggle it. |
/chronicle |
Session history tools and insights | Looks back over your past chats and turns them into useful summaries. Like a diary that writes itself. Options: standup (what you got done, ready to share)tips (personalized advice to use Copilot better)improve (suggests fixes to your project's "house rules" file)cost-tips (ways to use fewer tokens / save money)reindex (rebuild the history if subcommands say "no sessions found"). |
| Command | What Copilot shows you | In plain words (+ how to type it) |
|---|---|---|
/research |
Run deep research investigation using GitHub search and web sources | Does deep homework across GitHub and the web, with sources. Type /research best way to cache in Redis. ⭐ |
/ask |
Ask a quick side question without adding to conversation history | A quick side question that won't clutter the main chat. Type /ask what does this error mean? |
/search |
Search the conversation timeline | Find something said earlier in this chat without scrolling. Like Ctrl+F for your conversation. Type /search then a word, e.g. /search database. |
| Command | What Copilot shows you | In plain words (+ how to type it) |
|---|---|---|
/voice |
Manage voice mode (dictation transcription via Foundry Local) | Talk instead of type: your speech becomes text on your own device (private). First use downloads a model; pick the larger one for fewer errors. |
/help |
Show help for interactive commands | The full command list. Type /help. |
/changelog |
Display changelog for CLI versions. Add 'summarize' to get an AI summary. | See what's new in each version of Copilot CLI. Options: type /changelog for the list, or /changelog summarize to get a short AI summary instead of the raw notes. |
/feedback |
Provide feedback about the CLI | Tell the GitHub team what's good or broken. Type /feedback. |
/theme |
View or set color mode | Changes the colors of the CLI to suit your eyes or your room's lighting. Type /theme and pick light, dark, or a custom look. |
/statusline |
Configure status line items | Customizes the little info bar so it shows the details you care about. Type /statusline to choose what appears. |
/footer |
Configure status line items | Same idea as /statusline: tweak what shows in the bottom bar. Type /footer. |
/streamer-mode |
Toggle streamer mode (hides preview model names and quota details for streaming) | Hides private details when screen-sharing: your quota % in the footer, and preview model names (in the /model list they show as "Hidden Model 1, 2, 3…"). Options: on (hide)off (show again)models (reveal just the model names). Turn it ON before you present. |
/instructions 🔧 |
View and toggle custom instruction files | Shows the "house rules" files Copilot is following, and lets you turn them on or off. Type /instructions. |
/env 🔧 |
Show loaded environment details (instructions, MCP servers, skills, agents, plugins, LSPs, extensions) | One screen showing everything that's currently loaded: all your instructions, connected tools, skills, and plugins in one place. Great for "wait, what's switched on right now?" Type /env. |
/experimental 🔧 |
Show available experimental features, or enable/disable experimental mode | Turn on early, in-progress features. Type /experimental. |
| Command | What Copilot shows you | In plain words (+ how to type it) |
|---|---|---|
/login |
Log in to Copilot | Sign in with your GitHub account. Type /login. |
/logout |
Log out of an OAuth login session | Sign out. Type /logout. |
/user |
Manage GitHub user list | Switch between GitHub accounts (handy if you have a work and a personal one). Options: show (current user)list (all added users)switch (change account). |
/new |
Start a new conversation | Fresh chat, same session. Type /new. |
/clear |
Abandon this session and start fresh | Throw away this session and start over. Type /clear. |
/restart |
Restart the CLI, preserving the current session | Reboot the app but keep your work. Type /restart. |
/update |
Update the CLI to the latest version | Updates Copilot CLI itself (runs npm i -g @github/copilot). Add prerelease to get early test versions. Note: this does NOT update your MCP servers, skills, or plugins; those are managed separately via /mcp, /skills, /plugin. |
/version |
Display version information and check for updates | Show which version you're on. Type /version. |
/exit |
Exit the CLI; use 'print' to print the session after exiting alt screen | Quits Copilot CLI. Type /exit. Add print (/exit print) to leave the conversation printed in your terminal instead of clearing the screen. |
(⭐ 🔧
These special keys work right in the prompt:
| Type this | What it does | In plain words |
|---|---|---|
@ |
Mention files | Attach a file by name so Copilot can read it. |
# |
Mention issues and pull requests | Point Copilot at a GitHub issue or PR. |
! |
Execute shell command | Run a terminal command directly (e.g. !ls -la). |
/ |
Commands | Opens the slash-command menu (everything above). |
| Keys | Action |
|---|---|
shift+tab |
Switch modes (chat / agent / autopilot) |
ctrl+s |
Run command, keep your typed input |
ctrl+o / ctrl+e |
Expand all timeline entries |
ctrl+c |
Cancel the current action |
ctrl+c ×2 |
Exit |
esc |
Cancel |
ctrl+d |
Shutdown |
ctrl+l |
Clear the screen |
ctrl+t |
Toggle the reasoning display |
ctrl+x → b |
Move the current task to the background |
ctrl+x → o |
Open the most recent link |
| Keys | Action |
|---|---|
ctrl+a / ctrl+e |
Jump to start / end of line |
ctrl+h |
Delete the previous character |
ctrl+w |
Delete the previous word |
ctrl+u / ctrl+k |
Delete to start / end of line |
meta+← / meta+→ |
Move by word |
ctrl+g |
Edit your prompt in $EDITOR |
Copilot CLI reads all of these files (the ones that exist), so it remembers how your project works:
CLAUDE.md ← your existing Claude file works as-is!
GEMINI.md
AGENTS.md ← in git root & current folder
.github/instructions/**/*.instructions.md
.github/copilot-instructions.md
$HOME/.copilot/copilot-instructions.md
COPILOT_CUSTOM_INSTRUCTIONS_DIRS ← env var for extra folders
Does it understand CLAUDE.md? Yes, no rename needed. It's a built-in convention; Copilot CLI looks for CLAUDE.md automatically.
Which one "wins" if I have several? None. They're merged, not ranked. Copilot loads every file that exists and combines them into one instruction set (e.g. CLAUDE.md plus AGENTS.md plus .github/copilot-instructions.md are all honored together). So your old Claude setup keeps working, and you can layer Copilot-specific rules on top.
⚠️ The one thing to avoid: don't write contradictory rules in two different files. Since all of them reach the model, conflicting instructions confuse the answer, not the file-loading. Keep your rules consistent across files.
Migration tip: your existing CLAUDE.md is honored as-is. No rewrite needed.
- It asks before acting. Every shell command asks for an OK by default. Use
/allow-all(or/add-dir) in folders you trust to speed things up. /delegateis magic when available. In an eligible GitHub repo it hands the task to the cloud and opens a pull request for you./fleetruns many agents at once. Great for big, repetitive jobs./remote= phone control. Walk away and steer from GitHub mobile or web./everyand/afterschedule work. "Every 10 minutes, run the tests." It's a cron job for your agent.
START copilot
HELP /help
PICK A BRAIN /model (Claude / GPT-5 / Gemini …)
ATTACH FILE @path/to/file
ISSUE / PR #123
RUN SHELL !ls -la
SWITCH MODES shift+tab
TALK TO IT /voice
UNDO /undo or /rewind
NEW CHAT /new
TRUST A FOLDER /allow-all (only in folders you trust!)
BIG TASK /delegate (→ GitHub PR)
MANY AT ONCE /fleet
SECOND OPINION /rubber-duck
RESEARCH /research <topic>
SCHEDULE /every 10m <prompt> · /after 30m <prompt>
PHONE CONTROL /remote on
QUIT /exit
Welcome to Copilot CLI. Same brain if you want it, a bigger toolbox, and deeper GitHub. 🚀