GitHub Copilot CLI v1.0.60 Changelog
What changed in Copilot CLI v1.0.60, released 2026-06-05 β explained in plain English.
- You can now create "git worktrees" (extra checkouts of your code) straight from pull requests, and trust settings carry over to them automatically.
- The AI is smarter about saving memory: custom agent instructions aren't duplicated, and you can use max reasoning effort on any plan.
- Tons of visual bugs are fixed, like screens freezing after sleep, broken copy-paste on Linux, and weird cursor positions.
Want a heads-up when GitHub Copilot CLI updates?
We email you the moment a new GitHub Copilot CLI version drops β plus plain-English release notes like these. No spam.
This site is protected by reCAPTCHA and the Google Privacy Policy and Terms of Service apply.
This release is all about making the terminal experience smoother and safer for developers who live in the command line. It matters most to power users who juggle multiple coding sessions, use AI agents to write code, and need a reliable, glitch-free interface.
What's new
- Create worktrees from PRs: A "git worktree" lets you check out a branch in a separate folder without messing up your current workspace. You can now set one up directly from the pull request screen.
- Max reasoning effort for all: When the AI thinks through a problem, you can now crank its "reasoning effort" up to the maximum level, no matter what pricing plan you're on.
- Rubber duck agent setting: There's a new setting (
builtInAgents.rubberDuckAutoInvoke) to let an AI "rubber duck" automatically help you talk through your code. It's turned off by default. - Stash your prompts: Press
Ctrl+Sto temporarily save (stash) what you're typing so you can do something else, and pop it back later. - Vim-style diff navigation: If you like Vim keyboard shortcuts, you can now use keys like
g,G,Ctrl+D, andCtrl+Uto move around the/diff(code comparison) view. - Auto-link issues: If you type a bare hashtag number like
#42in your prompt, it automatically links to issue or PR #42 in your current Git repository. - Track AI token usage: The
/usagecommand now shows "cache write tokens" alongside "cache read tokens," helping you see exactly how much AI memory you're using. - Context and billing details: The
/contextcommand now separates your custom instructions from the main system prompt and shows token costs for MCP (Model Context Protocol) tools. There's also a newbillinghelp topic to explain your AI credits. - Shorthand flags: You can now type
-rinstead of--resumeto quickly jump back into a previous session. - Security prompts for sensitive commands: The tool will now ask for your permission before running commands that change dynamic loaders or Git configs (like
LD_PRELOADorGIT_EXTERNAL_DIFF), keeping your system safe from sneaky overrides. - Web fetch safety: The
web_fetchtool now blocks access to private, local, and cloud metadata IP addresses, and it won't silently follow web redirects. This prevents accidental data leaks.
Improvements
- Better narrow screens: If you shrink your terminal window, the status line now neatly stacks information across multiple rows instead of cutting it off into gibberish.
- Smaller model picker: The menu for picking AI models now fits properly inside small terminal windows, and you can scroll through it with your mouse.
- Smarter IDE picker: If an IDE (code editor) disconnects, it hides from the menu so you don't accidentally click it and get an error. If you have multiple windows open for the same folder, it adds a process ID number so you can tell them apart.
- Rewind stats: The "Rewind" feature (which lets you go back in time) now shows a
+added βremovedcount of your code changes at each checkpoint. - Less memory waste: Custom agent instructions are no longer copied and pasted every single turn, which saves your AI's context window (its short-term memory).
- Windows executable safety: On Windows, if you type a bare command like
git, it will no longer accidentally run a random executable sitting in your current folder. It only runs trusted system commands unless you specifically add your folder to thePATH. - Cleaner menus: Multi-line descriptions in skill and slash-command menus now display as a single, readable line.
- Helpful help menu: Typing "help" now opens a quick-help overlay instead of sending the word "help" as a prompt to the AI. Missing shortcuts were also added to the
/helpguide. - Environment tracking: The
/envcommand now shows how many "hooks" (custom automated scripts) are running and where they came from. - Session sharing: You can now see your Mission Control sharing status and session links when you look at your session info.
Fixes
- Sleep freeze fixed: The screen no longer stays blank when waking up from sleep inside a terminal multiplexer (a tool that lets you split terminal windows).
- Clipboard fix for Linux: Copying and pasting on X11 (a common Linux display system) no longer corrupts or breaks your terminal display.
- Cursor position fixed: The blinking cursor now shows up in the right spot when you're approving plans or giving feedback.
- Input field colors fixed: Input boxes now correctly show their background colors inside highlighted frames.
- No more hanging commands: Interactive shell commands no longer freeze up when they spit out a massive amount of text all at once.
- Wide character fix: Wide characters (like Japanese, Chinese, or Korean text) now render perfectly in the code comparison view without visual glitches.
- Rewind file protection: Using Rewind to go back to an older snapshot no longer accidentally deletes your ignored files (like local
.envfiles). - Negative limits fixed: If you go over your usage limit, the screen no longer incorrectly shows a negative percentage for remaining requests.
- Linux sandbox fix: The Linux sandbox no longer crashes when you set up
allowedHostsorblockedHostsrules. - Background command wait: The terminal beep and autopilot signals now wait for your background shell commands to finish before firing off.
- macOS delete shortcut:
Cmd+Backspacenow correctly deletes the line *before* your cursor on macOS. - Config and plugin fixes: Trusted folders are no longer forgotten when experiments are cached. Force-removing a marketplace no longer makes its plugins reinstall. Repository plugin overrides no longer mess up your global plugin settings.
- MCP server fixes: Several bugs with MCP servers (external tools the AI can use) were squashed. OAuth logins no longer fail if already in progress, local servers start reliably, and servers registered via Azure API Center are no longer wrongly blocked.
- LSP server fix: The LSP (Language Server Protocol) now return the correct number of entries, stopping strict servers from panicking and crashing.
- Symlinked extensions: Extensions linked via directory shortcuts (symlinks) are now found and loaded properly.
- BYOK large files: If you bring your own AI API key (BYOK), file attachments larger than 5 MiB now send successfully through OpenAI without failing.
- Git init prompt removed: The tool no longer annoyingly suggests running
/initwhen you are working outside of a Git repository. - Tab traversal fix: Hitting
Tabwhile typing..to navigate up a folder path now correctly moves up the path instead of switching your tabs. - Queue hint fix: The keyboard hint now correctly shows
ctrl+enterinstead ofctrl+qwhen using the kitty keyboard protocol. - MCP glyph size: The little icon for MCP tools in the
/contextlegend is now the correct size. - Worktree naming: Worktree folders now use a flat name with dashes when a branch has slashes (so
cli/foobecomescli-foo), preventing folder path issues.
Breaking changes
- Windows bare commands: As mentioned in the improvements, Windows will no longer run executables from your current folder when you type a bare name like
git. If you relied on this behavior, you must add your working directory to your system'sPATHvariable to turn it back on.
How to upgrade
Update to GitHub Copilot CLI v1.0.60 by running:
npm install -g @github/copilot
This is our plain-language summary. Read the complete, official notes on the GitHub Copilot CLI official changelog β.
