No entries match these filters. Clear filters
advisory CVE-2026-102230
Nov 10, 2026
·
issue 139
CVE-2026-102230: Information Disclosure via Integrated Browser File Uploads (CVSS CVSS Common Vulnerability Scoring System, a method for describing a vulnerability's characteristics and severity. Its score is not the probability that attackers will exploit it. 7.4, affects < 1.145.3)︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet
advisory CVE-2026-121662
Nov 10, 2026
·
issue 139
CVE-2026-121662: Workspace Trust Bypass via Prototype Pollution in Workspace Configuration (CVSS CVSS Common Vulnerability Scoring System, a method for describing a vulnerability's characteristics and severity. Its score is not the probability that attackers will exploit it. 8.6, affects < 1.145.3)︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet
advisory CVE-2026-121660
Nov 10, 2026
·
issue 139
CVE-2026-121660: Security Feature Bypass via Inherited Publisher Names in Allowed Extensions Policy (CVSS CVSS Common Vulnerability Scoring System, a method for describing a vulnerability's characteristics and severity. Its score is not the probability that attackers will exploit it. 7.5, affects < 1.145.3)︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet
advisory CVE-2026-121664
Nov 10, 2026
·
issue 139
CVE-2026-121664: Remote Code Execution via Extension Text Rendered as Markdown (affects < 1.145.3)︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet
advisory CVE-2026-119975
Nov 10, 2026
·
issue 139
CVE-2026-119975: Command Injection via Crafted Publisher Names in the Extension Trust Dialog (affects < 1.145.3)︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet
advisory CVE-2026-120782
Nov 10, 2026
·
issue 139
CVE-2026-120782: Command Execution via Agent-Originated Local Commands (CVSS CVSS Common Vulnerability Scoring System, a method for describing a vulnerability's characteristics and severity. Its score is not the probability that attackers will exploit it. 6.8, affects < 1.145.3)︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet
advisory CVE-2026-120780
Nov 10, 2026
·
issue 139
CVE-2026-120780: Remote Code Execution via serve-web Release Downgrade (CVSS CVSS Common Vulnerability Scoring System, a method for describing a vulnerability's characteristics and severity. Its score is not the probability that attackers will exploit it. 7.2, affects < 1.145.3)︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet
code-oss adds enterprise-managed MCP MCP Model Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. server allowlists and denylist controls, with registry precedence rules starting in VS Code 1.145︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ USE IT
Allow only MCP servers under a corporate domain while blocking a specific endpoint — deliver this through managed settings so developers cannot override it in their editor or mcp.json.
json
copy
{
"allowedMcpServers": [
{ "serverUrl": "https://mcp.example.com/*" }
],
"deniedMcpServers": [
{ "serverUrl": "https://mcp.example.com/blocked" }
]
}
Prevent all MCP server access across the fleet — an explicit empty allowlist blocks every server, including those in the registry.
json
copy
{
"allowedMcpServers": []
}
Lock the Copilot harness and disallow Local, Claude, and Codex for all managed users — set this in enterprise-managed settings to take precedence over any user-side default.
json
copy
{
"vscode.temporaryAllowLocalHarness": false
}
Control which MCP servers developers can install and run by setting allowedMcpServers in Copilot enterprise-managed settings; each rule uses one of three matchers — serverUrl (supports * wildcards), serverCommand (exact command plus arguments), or serverName — delivered through a supported managed-settings method, not in a developer's editor settings or mcp.json.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Block specific MCP servers even when they match the allowlist by setting deniedMcpServers in enterprise-managed settings; deny rules and explicit MCP disablement such as chat.mcp.access set to 'none' take priority over the allowlist. Not checked yet Starting with VS Code 1.145, an enterprise-managed allowlist takes precedence over registry-only access for editor-managed MCP servers: a server matching the allowlist but absent from the registry is permitted, while a server in the registry that does not match the allowlist is blocked. Not checked yet Starting with VS Code 1.145, a temporary network or service failure while refreshing GitHub organization or enterprise MCP registry policy no longer removes a previously received restriction — VS Code retains the last known registry URL and access restriction; configure the ChatMCP device policy for restrictions that must apply independently of a successful GitHub policy fetch. Not checked yet Prohibit the Local, Claude, and Codex harnesses by setting vscode.temporaryAllowLocalHarness to false in enterprise-managed settings (available from VS Code 1.143); this takes precedence over vscode.temporaryDefaultToLocalHarness: true. Not checked yet
+2 more
show less
Prefer the Copilot harness for new editor chats without prohibiting other harnesses by enabling the ChatEditorPreferCopilotHarness device policy. Not checked yet Enforce supported minimum sandbox restrictions by using the shared sandbox key in enterprise-managed settings, following the sandbox schema and minimum client versions. Not checked yet
7 details in Everything
code-oss adds three new Copilot policy keys for controlling Canvases, Connectors, and the customization marketplace︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
Control whether Copilot sessions can present interactive content by setting the chat.canvases.enabled policy (ChatCanvases).︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Enable Copilot Connectors in the marketplace and expose connected MCP MCP Model Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. servers in Chat Customizations and agent sessions with the chat.customizations.copilotConnectors.enabled policy (ChatCopilotConnectors). Not checked yet Switch the customization marketplace to a Browse view instead of Overview when a marketplace source is enabled, using the chat.customizations.marketplace.enabled policy (ChatCustomizationMarketplace). Not checked yet
3 details in Everything
code-oss adds remote-proxy browsing in the integrated browser, letting you preview apps on SSH, WSL WSL Windows Subsystem for Linux, a Microsoft-built compatibility layer that lets a Linux environment run directly on Windows without a virtual machine, giving cyber tools access to Linux binaries, filesystems, and syscalls on a Windows host. , Dev Container Dev Container A Docker-based development environment defined in a JSON config file, standardized by Microsoft, that lets a project ship its exact toolchain so any contributor's editor or CI system gets an identical workspace. , and dev-tunnel hosts without forwarding a port︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ HOW TO FIND IT
Preview a web app running inside a Dev Container or SSH host without setting up port forwarding.
📍 1. In VS Code, connect to a remote agent host (SSH, WSL, Dev Container, or dev tunnel) and open a project folder.
2. Start your web app on the remote host (e.g. on port 3000).
3. Open the integrated browser and navigate to http://localhost:3000.
4. Confirm the remote indicator in the address bar identifies the session's host — traffic is now proxied through the remote connection.
5. To let the agent interact with the app, select the share option in the browser toolbar and confirm sharing.
The integrated browser now proxies traffic through a remote workspace connection (Dev Container, SSH host, WSL, or GitHub Codespace), so you can preview apps on the remote machine without forwarding a port first.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Agent-opened pages use separate, in-memory storage per chat by default, while tabs you open with default storage persist cookies and localStorage across sessions for the same project folder and host, including linked worktrees. Not checked yet
2 details in Everything
code-oss adds agent plugin management with /plugin commands, marketplace configuration via chat.plugins.marketplaces, and local registration via chat.pluginLocations︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ USE IT
Pin a private org marketplace to an exact commit for reproducible plugin installs across a team.
settings.json
copy
// settings.json
{
"chat.plugins.marketplaces": [
"your-org/plugin-marketplace#a1b2c3d4e5f6a1b2c3d4e5f6a1b2c3d4e5f6a1b2"
]
}
Register a plugin under active development locally without a full install, so changes are picked up immediately.
settings.json
copy
// settings.json
{
"chat.pluginLocations": {
"/path/to/my-plugin": true
}
}
Recommend a shared internal plugin to everyone who opens the repo, using the qualified marketplace identity.
.github/copilot/settings.json
copy
// .github/copilot/settings.json
{
"extraKnownMarketplaces": [
{
"company-tools": {
"source": {
"source": "github",
"repo": "your-org/plugin-marketplace"
}
}
}
],
"enabledPlugins": [
"code-formatter@company-tools"
]
}
Configure additional marketplace repositories using the chat.plugins.marketplaces setting in settings.json, supporting shorthand owner/repo, HTTPS HTTPS HTTP carried over an encrypted TLS connection. It protects traffic in transit and lets a browser verify the server's certificate. .git URLs, SCP SCP Secure Copy Protocol, a file-transfer method built on SSH that encrypts data in transit; cyber tools use it to move files to or from remote hosts without exposing credentials or content. -style SSH remotes, and file:/// local paths — with #<ref> to pin a branch, tag, or full 40-character commit SHA for reproducible installs.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Register a local plugin development directory without installing it using chat.pluginLocations in settings.json, mapping directory paths to true (enabled) or false (registered but disabled). Not checked yet Configure workspace-level plugin recommendations in .claude/settings.json or .github/copilot/settings.json using extraKnownMarketplaces to register additional marketplaces and enabledPlugins to list plugins enabled by default, with plugin identifiers in <plugin>@<marketplace> format. Not checked yet Manage plugins from a Copilot harness session with /plugin commands: /plugin list, /plugin install <source>, /plugin update <plugin>, /plugin enable <plugin>, /plugin disable <plugin>, /plugin uninstall <plugin>. Not checked yet Manage marketplaces from a Copilot harness session with /plugin marketplace subcommands: list, browse <name>, add <source>, update [name], and remove <name> (with --force to remove even when plugins from that marketplace are installed). Not checked yet
+2 more
show less
Attach plugin-wide lifecycle hooks that run shell commands at specific points in the agent workflow, firing alongside workspace and user-level hooks whenever the plugin is enabled. Not checked yet Schedule plugin hooks on a manual, hourly, or five-field cron schedule — for example "0 9,17 * * MON-FRI" — supporting lists, ranges, and steps; named time zones are not supported and invalid templates are silently ignored. Not checked yet
7 details in Everything
code-oss adds organization-level custom agents, user-level agent folders, and new frontmatter and settings for custom agent discovery and hooks︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ USE IT
Disable org-level custom agents when you want to suppress shared organization agents from appearing in your chat dropdown.
.vscode/settings.json
copy
"github.copilot.chat.organizationCustomAgents.enabled": false
Enable parent-repository agent discovery so a monorepo's root-level agents are available in all child workspaces.
.vscode/settings.json
copy
"chat.useCustomizationsInParentRepositories": true
Disable organization-level custom agents in chat by setting github.copilot.chat.organizationCustomAgents.enabled to false.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Enable custom agent discovery from a monorepo's parent repository root with the chat.useCustomizationsInParentRepositories setting. Not checked yet Store workspace-scoped agents as .agent.md files in the .github/agents folder. Not checked yet Configure additional agent file search locations with the chat.agentFilesLocations setting, which remains supported across harnesses including Copilot. Not checked yet Create a new custom agent with the Chat: New Custom Agent command, which stores the file in a harness-supported location via the Agent Customizations editor. Not checked yet
+2 more
show less
code-oss now supports organization-level custom agents, automatically detected and shown alongside built-in, personal, and workspace agents in the agents dropdown. Not checked yet Use hooks in agent frontmatter to run automation at configured points in an agent's workflow; VS Code preserves hooks when converting agents for Copilot, including plugin-provided agents. Not checked yet
Disable organization-level custom agents in chat by setting github.copilot.chat.organizationCustomAgents.enabled to false.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Enable custom agent discovery from a monorepo's parent repository root with the chat.useCustomizationsInParentRepositories setting. Not checked yet Store workspace-scoped agents as .agent.md files in the .github/agents folder. Not checked yet Create a new custom agent with the Chat: New Custom Agent command, which stores the file in a harness-supported location via the Agent Customizations editor. Not checked yet code-oss now supports organization-level custom agents, automatically detected and shown alongside built-in, personal, and workspace agents in the agents dropdown. Not checked yet
+1 more
show less
Use hooks in agent frontmatter to run automation at configured points in an agent's workflow; VS Code preserves hooks when converting agents for Copilot, including plugin-provided agents. Not checked yet
1 more in Everything
└──▷ BREAKING ON UPGRADE
!
The `chat.modeFilesLocations` setting is deprecated; migrate to `chat.agentFilesLocations` or native harness folders.
Our best guess
!
`.chatmode.md` files must be renamed to `.agent.md` and placed in a supported custom agent location.
Our best guess
!
Profile-storage agents (from the local harness) are not automatically available to other harnesses; move them manually to the target harness's user folder (e.g., `~/.copilot/agents` or `~/.claude/agents`).
Our best guess
code-oss adds mixed-harness chats, local memory storage in .github/copilot-memories.jsonl, and new settings to control reasoning and tool-call display during persistent pro︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ USE IT
Store Copilot repository memories locally so they can be reviewed and committed alongside code rather than saved to GitHub cloud storage.
.vscode/settings.json
copy
// In VS Code settings (settings.json)
{
"chat.copilot.memory.local.enabled": true
}
// Memories will be written to .github/copilot-memories.jsonl in the repository root.
Run multiple chats in one session using different harnesses, each with its own model and permissions, to compare Copilot Local and cloud agent responses side by side.
.vscode/settings.json
copy
// In VS Code settings (settings.json)
{
"chat.agentHost.multipleChatProviders.enabled": true,
"chat.agentHost.moveChats.enabled": true
}
Enable mixed-harness chats in one session via chat.agentHost.multipleChatProviders.enabled, letting each chat use its own model and permission settings in the Agents window.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Move chats between sessions or into a new session in the Agents window using chat.agentHost.moveChats.enabled (also requires chat.agentHost.multipleChatProviders.enabled to move between harnesses). Not checked yet Store repository memories locally in .github/copilot-memories.jsonl instead of the cloud by enabling chat.copilot.memory.local.enabled; the file can be committed, and cloud-memory policies do not apply. Not checked yet Control reasoning and tool-call display during persistent progress with chat.persistentProgress and chat.persistentProgressVerbosity, which can preview running tool calls, keep tool calls expanded, or group reasoning into collapsed sections. Not checked yet Show a separate Environment picker for new sessions in the Agents window (requires the unified workspace picker); when off, environment choices remain in the workspace and harness pickers. Not checked yet
5 details in Everything
code-oss adds external ingest for semantic workspace indexing on non-GitHub repositories, controlled by github.copilot.chat.workspace.codeSearchExternalIngest.enabled︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ USE IT
Enable external ingest for a local or non-GitHub workspace so Copilot can run semantic search semantic search Searching by meaning rather than requiring the same words. It commonly compares vector representations to find related content, even when the wording differs. over it.
.vscode/settings.json
copy
github.copilot.chat.workspace.codeSearchExternalIngest.enabled: true
Enable semantic search on local folders and non-GitHub repositories by turning on github.copilot.chat.workspace.codeSearchExternalIngest.enabled in workspace settings — this activates external ingest, which uploads indexable files to GitHub to make them searchable.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Rebuild or build the semantic index on demand for the current workspace using the Codebase Semantic Index command from the VS Code Command Palette. Not checked yet Check index availability, policy blocks, and external ingest status through the Copilot status dashboard in the VS Code Status Bar under Codebase Semantic Index . Not checked yet When a semantic index is unavailable, agents now fall back automatically to text search, grep, file search, and language intelligence without requiring any index. Not checked yet
Enable semantic search on local folders and non-GitHub repositories by turning on github.copilot.chat.workspace.codeSearchExternalIngest.enabled in workspace settings — this activates external ingest, which uploads indexable files to GitHub to make them searchable.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Rebuild or build the semantic index on demand for the current workspace using the Codebase Semantic Index command from the VS Code Command Palette. Not checked yet Check index availability, policy blocks, and external ingest status through the Copilot status dashboard in the VS Code Status Bar under Codebase Semantic Index . Not checked yet
1 more in Everything
code-oss adds remote Dev Container Dev Container A Docker-based development environment defined in a JSON config file, standardized by Microsoft, that lets a project ship its exact toolchain so any contributor's editor or CI system gets an identical workspace. support for agents and an integrated browser preview for remote web apps in VS Code 1.139︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ USE IT
Enable the remote browser proxy so an agent can preview and interact with a web app running on a remote host without forwarding its port.
yaml
copy
workbench.browser.enableRemoteProxy: true
workbench.browser.dataStorage: default
Run agents inside a remote project's Dev Container for SSH, Tunnel, and WSL WSL Windows Subsystem for Linux, a Microsoft-built compatibility layer that lets a Linux environment run directly on Windows without a virtual machine, giving cyber tools access to Linux binaries, filesystems, and syscalls on a Windows host. connections in the desktop Agents window, available starting in VS Code 1.139.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Preview a remote web app without port-forwarding by enabling workbench.browser.enableRemoteProxy, leaving workbench.browser.dataStorage at default, and opening the app's original localhost URL in the integrated browser. Not checked yet Use Share with Agent in the integrated browser to let the agent interact with a running remote web app or validate a user flow. Not checked yet
3 details in Everything
code-oss adds a Customization Marketplace for MCP MCP Model Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. servers, a Session Target dropdown, MCP: Browse MCP Servers command, and chat.customizations.marketplace.enabled setting.
Control the new discovery flow vs. the legacy Extensions gallery with the chat.customizations.marketplace.enabled setting (default: on). Not checked yet New integration Run MCP: Browse MCP Servers from the Command Palette to open MCP server discovery in the Agent Customizations editor. Not checked yet Select the Session Target dropdown to choose the destination harness before adding a server. Not checked yet Use automatic MCP server discovery to import an existing MCP configuration from another application. Not checked yet
4 details in Everything
code-oss adds custom agent authoring with .github/agents/<name>.agent.md files, plugin reuse from marketplaces, handoffs, and an Agent Customizations editor︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ TRY IT
Create a code-review agent scoped to your workspace so teammates can select it immediately from the agent dropdown.
copy
$ # Save as .github/agents/reviewer.agent.md in your project root
# (create the folders if they don't exist)
mkdir -p .github/agents
cat > .github/agents/reviewer.agent.md << 'EOF'
---
name: Reviewer
description: Review code for incorrect behavior and missing tests without editing files
tools: [read, search]
---
Review the code supplied in the request. Read related code and tests
when needed to check the behavior.
Focus on:
- Incorrect behavior and unhandled errors.
- Missing tests for important behavior and error cases.
Report only actionable findings. For each finding, include the file
and line, the problem, its impact, and a suggested fix or test.
If you find no issues, say so and note anything you couldn't verify.
Do not edit files or run terminal commands.
EOF
Save a custom agent as .github/agents/reviewer.agent.md (or any name) using a YAML frontmatter header for name and tools and a Markdown body for instructions, then select it from the agent dropdown.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Open the Agent Customizations editor with the Agents window or the agent dropdown to list, locate, edit, and delete custom agents, and to create new ones with the plus icon. Not checked yet Browse and install agent plugins from customization marketplaces before authoring your own, then manage installations through the Plugins section. Not checked yet Connect agents into a workflow with handoffs — after a response completes, handoff buttons let you switch to another agent with conversation context and a pre-filled prompt; custom-agent handoffs are supported only in the Local harness. Not checked yet Share a workspace agent with teammates by committing .github/agents/<name>.agent.md to source control; teammates on a compatible harness can select it from the agent dropdown. Not checked yet
+1 more
show less
Distribute agents alongside skills, instructions, or tools as an agent plugin. Not checked yet
6 details in Everything
code-oss adds agent plugin deep links, remote plugin support, and a Customization Marketplace for discovering and installing Copilot agent plugins︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ TRY IT
Share a one-click install link for an agent plugin in a README badge or web page, so teammates can install it directly into VS Code.
copy
$ https://vscode.dev/redirect?url=vscode%3A%2F%2Fchat-plugin%2Finstall%3Fsource%3Downer%252Frepo%26plugin%3Dplugin-name
Register a plugin directory that lives on a remote machine (SSH, WSL WSL Windows Subsystem for Linux, a Microsoft-built compatibility layer that lets a Linux environment run directly on Windows without a virtual machine, giving cyber tools access to Linux binaries, filesystems, and syscalls on a Windows host. , Dev Containers Dev Containers A Docker-based development environment specification, maintained by Microsoft and the open-source community, that lets a tool ship a fully configured, reproducible workspace any developer can launch inside VS Code or a compatible editor. ) without copying it locally.
📍 1. Open the Command Palette in the remote window and run Preferences: Open Remote Settings (JSON) .
2. Add the Experimental local plugin registration setting with the plugin path on the remote filesystem.
3. Start a new chat with the Copilot harness and verify the plugin's components load.
Install a specific agent plugin via the vscode://chat-plugin/install?source=<url-encoded-source>&plugin=<url-encoded-plugin-name> deep link, which registers the marketplace and installs the named plugin in one step.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Register a plugin marketplace without installing any plugins via the vscode://chat-plugin/add-marketplace?ref=<url-encoded-source> deep link. Not checked yet Use the HTTPS HTTPS HTTP carried over an encrypted TLS connection. It protects traffic in transit and lets a browser verify the server's certificate. redirect wrapper https://vscode.dev/redirect?url=vscode%3A%2F%2Fchat-plugin%2Finstall%3Fsource%3Downer%252Frepo%26plugin%3Dplugin-name to publish plugin install links on web pages or markdown badges. Not checked yet Use plugins on remote machines (Remote - SSH, WSL, Dev Containers, Remote - Tunnels) by registering a plugin directory via Preferences: Open Remote Settings (JSON) and the Experimental local plugin registration setting. Not checked yet Enter slash commands in a Copilot harness session to view completions for plugin and marketplace operations as an alternative to the Agent Customizations editor. Not checked yet
5 details in Everything
code-oss adds an Agent Customizations editor and vscode://chat-customizations/open deep links for navigating agent customization sections directly︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ TRY IT
Jump straight to the skills section pre-filtered to Python customisations — useful when onboarding a new Python project and looking for reusable skills.
copy
$ vscode://chat-customizations/open?page=skills&search=python
Share a markdown badge or link in a README that opens the plugin marketplace so contributors can install team plugins in one click.
copy
$ https://vscode.dev/redirect?url=vscode%3A%2F%2Fchat-customizations%2Fopen%3Fpage%3Dplugins
Open the Customization Marketplace discover page to browse available integrations before creating a custom skill from scratch.
📍 Steps: Select the intended harness from the Session Target control in a new chat › Open the Agent Customizations editor › Go to the Discover section to browse the Customization Marketplace.
Navigate directly to any section of the Agent Customizations editor using vscode://chat-customizations/open?page=<page>&search=<query> deep links; supported page values include discover, agents, skills, instructions, hooks, mcp-servers (or mcp-server), plugins, tools, and migrations; use vscode-insiders:// for Insiders builds or https://vscode.dev/redirect?url=<url-encoded-vscode-URL> for web-safe sharing.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Adds a Customization Marketplace Preview that brings together plugin marketplaces, GitHub Feed, and Copilot connectors into a single discover page for finding skills, plugins, MCP MCP Model Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. servers, and service integrations without building from scratch. Not checked yet Adds an Agent Customizations editor that organises customisation types — custom agents, skills, instructions, hooks, MCP servers, plugins, and tools — into separate sections with configuration forms, inline file editors, or read-only details depending on type and source. Not checked yet Adds a Chat Customizations Evaluations extension that analyses skills, custom agents, instructions, and prompt files for ambiguity, contradictions, and overly complex rules, surfacing diagnostics before testing on a representative task. Not checked yet
4 details in Everything
code-oss adds mixed-harness chat sessions, harness visibility on nested chats, and policy-driven auto-migration from Local to Copilot harness︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ USE IT
Prevent the Local, Claude, and Codex harnesses from being used organization-wide starting at version 1.143, overriding any temporary allowance.
yaml
copy
vscode.temporaryAllowLocalHarness: false
Group chats that use different harnesses in a single session with mixed-harness chats, which starts a separate conversation on the chosen harness rather than handing off an existing one.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet See which harness a nested chat uses by hovering over its chat tab or its entry in the sessions list when it differs from the parent session's harness. Not checked yet Runtime policy can now automatically move unarchived Local harness conversations to the Copilot harness on open, preserving history and unsent input where conversation import is supported, or attaching the transcript as context via Continue in Copilot where it is not. Not checked yet Canvases are now supported in remote Agent Host sessions in the Agents window for Copilot sessions. Not checked yet Use BYOK BYOK Bring Your Own Key — a cloud deployment model where the customer supplies and controls the encryption keys rather than the provider, so sensitive data stays under the customer's cryptographic authority. credentials to access Claude-backed models in the Agents window while signed out of GitHub; Copilot-routed models become available after sign-in. Not checked yet
5 details in Everything
code-oss adds Dev Container Dev Container A Docker-based development environment defined in a JSON config file, standardized by Microsoft, that lets a project ship its exact toolchain so any contributor's editor or CI system gets an identical workspace. sandboxing sandboxing Running code inside a restricted environment that limits what it can access or change. The isolation reduces the impact of mistakes or untrusted code, depending on the sandbox's controls. support and read access for attached files in sandboxed Copilot sessions︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
Run agent sandboxing inside a Dev Container: when sandboxing is on, VS Code automatically starts the container with --security-opt seccomp=unconfined, --security-opt apparmor=unconfined, --security-opt systempaths=unconfined, and --device /dev/net/tun so the sandbox's namespaces and network access work correctly.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet In Copilot harness sessions with sandboxing enabled, files and folders explicitly attached to a message now receive read-only access for the session, without requiring an additional outside-workspace read approval in Manual permissions mode. Not checked yet
Run agent sandboxing inside a Dev Container: when sandboxing is on, VS Code automatically starts the container with --security-opt seccomp=unconfined, --security-opt apparmor=unconfined, --security-opt systempaths=unconfined, and --device /dev/net/tun so the sandbox's namespaces and network access work correctly.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet
1 more in Everything
code-oss adds scoped approval controls for terminal commands, web fetches, and file access in Copilot harness sessions︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ HOW TO FIND IT
Grant a persistent approval for a terminal command prefix so future Copilot harness sessions never re-prompt for it.
📍 In the Copilot harness Chat view, when a terminal command approval prompt appears for a command such as npm run build, open the approval dropdown and select Always Allow to save the prefix grant for future sessions on the same Agent Host.
Limit a file-read approval to the current session only, to avoid granting persistent access to sensitive paths.
📍 When a file read prompt appears in a Copilot harness session, open the approval dropdown and select in this Session so the grant expires when the session ends and is not saved to your user settings.
Scope approvals in Copilot harness sessions to reduce repeated prompts — choose Allow Once , in this Session , or Always Allow per tool call, covering a command prefix (such as npm run build), a host, a file, or a folder.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Terminal command approvals now sync with user settings via Always Allow , and web fetch approvals sync via Always Allow Fetching , without enabling response auto-approval. Not checked yet File and folder approvals are available as persistent grants — the folder scope covers the nearest parent Git repository, or the file's parent folder when no repository is found. Not checked yet Remote agent sessions keep their own approvals independently — local settings are not pushed to a remote host, and Always Allow on a remote host does not change local settings. Not checked yet Read-only or sandboxed terminal commands in Copilot harness sessions are approved by the runtime before approval rules are evaluated and do not prompt even when a matching rule or Always Allow is enabled. Not checked yet
5 details in Everything
└──▷ USE IT
Keep repository facts out of GitHub's cloud memory service and inside the repo's working tree — useful for air-gapped reviews or when org policy blocks cloud memory.
yaml
copy
chat.copilot.memory.local.enabled: true
code-oss adds mixed-harness nested sessions, chat-moving between sessions, and a worktree cleanup editor to the Agents window︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ HOW TO FIND IT
Allow running a Copilot chat and a Claude chat side-by-side in the same Agents session for implement-then-review workflows.
📍 In VS Code settings, search for chat.agentHost.multipleChatProviders.enabled and set it to true. Then open the Agents window, open a session, and use the harness picker above the input box to choose a different harness when creating a nested chat.
Reclaim disk space from idle agent session worktrees without losing branch history.
📍 In the Agents window, open the cleanup editor, choose an inactivity filter and optionally a workspace, review listed worktrees (branches, paths, estimated sizes), deselect any you want to keep, then run normal cleanup. Use Retry Failed Worktrees if any removals fail.
Move a nested chat to another session via 'Move to Session...' from the nested session's context menu, or move it to a new session via Move to New Session or by dragging it to empty space below the list. Not checked yet Adds a worktree cleanup editor to reclaim disk space from inactive or leftover session worktrees, with an inactivity filter, branch/path/size review, and Retry Failed Worktrees for failed removals. Not checked yet Nest a separate, interactive conversation under an existing session as a blank chat that shares the parent's harness and host by default. Not checked yet Use Force Remove in the cleanup editor to permanently discard uncommitted changes and remove a worktree used by other sessions or chats. Not checked yet
4 details in Everything
code-oss adds cron schedule support for local Agent Host automations with five-field expressions and time zone handling︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ HOW TO FIND IT
Run a commit-summarization automation twice each weekday without modifying files, using a cron schedule instead of a daily preset.
📍 1. Open your workspace in code-oss and navigate to the Agent Host automations panel.
2. Select 'Create a local Agent Host automation for this workspace'.
3. In the schedule picker, select Cron .
4. Enter the expression 0 9,17 * * MON-FRI across the five fields (or paste the full expression).
5. Confirm the schedule explanation shows 'At 9:00 AM and 5:00 PM, Monday through Friday' in your local time zone.
6. Set the automation prompt to: 'Summarize commits from the last 24 hours. Do not modify files.'
7. Review the first scheduled run, then enable the automation.
Run an automation on the first day of each month — a cadence no preset covers — using a cron expression.
📍 1. Open or edit a local automation and select Cron in the schedule picker.
2. Enter the expression '0 9 1 * *' to run at 9:00 AM on the first day of each month.
3. Confirm the schedule explanation and time zone, then save.
Schedule local Agent Host automations on a custom cadence by selecting Cron in the schedule picker and entering a five-field cron expression (minute, hour, day-of-month, month, day-of-week); presets like hourly, daily, and weekly remain available alongside the new cron option.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Cron expressions support * wildcards, comma-separated lists, ascending ranges such as MON-FRI, and positive steps on * or a range such as */30 or 0-59/15; steps from a single starting value like 5/15, a seconds field, and aliases such as @daily are not supported. Not checked yet Exported cron schedules use timeZone: local so the importing machine applies its own local time zone; a cron schedule whose saved time zone differs from the current local time zone cannot be exported. Not checked yet Cron schedules require a local Agent Host and are not supported for cloud automations. Not checked yet
4 details in Everything
code-oss's Agents window now supports Claude, ChatGPT-backed Codex, and BYOK BYOK Bring Your Own Key — a cloud deployment model where the customer supplies and controls the encryption keys rather than the provider, so sensitive data stays under the customer's cryptographic authority. models without requiring GitHub sign-in when provider credentials are configured︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
Run Claude, ChatGPT-backed Codex, and BYOK models in the desktop Agents window without GitHub sign-in by configuring supported provider credentials directly.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Focus the workspace and harness pickers in the Agents window via a keyboard shortcut, assignable through the keybinding configuration. Not checked yet
Run Claude, ChatGPT-backed Codex, and BYOK models in the desktop Agents window without GitHub sign-in by configuring supported provider credentials directly.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet
1 more in Everything
code-oss adds file-chooser attachment in chat to include files from outside the workspace︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ HOW TO FIND IT
Attach a UI mockup screenshot from your desktop to a chat prompt without opening a workspace.
📍 In the chat input, select the attachment icon, then choose Files , pick the image or text file from disk, select Attach , and type your prompt.
Attach files from disk — including text files and images — to a chat session via the file chooser (Files in the chat input) without needing to add the folder to the workspace first.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet In sessions that support only image attachments, Files automatically filters the chooser to supported image types. Not checked yet Attaching a file in a sandboxed Copilot harness session grants read access for that session without changing sandbox settings, write permission, or organization restrictions. Not checked yet
3 details in Everything
code-oss adds canvas support on remote agent hosts in the Agents window via workbench.browser.enableRemoteProxy︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ HOW TO FIND IT
Open a canvas served by an extension on a remote agent host — useful when your agent environment runs on a separate machine.
📍 1. In VS Code, open Settings and enable workbench.browser.enableRemoteProxy.
2. In the Agents window, connect to the remote host.
3. Start a new Copilot harness session and send a message.
4. Ask the agent to open a canvas provided by an extension installed on that host.
Canvases now work in the VS Code desktop app in Copilot harness sessions, including sessions hosted on your own machine. Not checked yet
1 detail in Everything
└──▷ HOW TO FIND IT
Keep tool calls visible as they complete instead of collapsing, so you can audit every step during an agentic chat session.
📍 More Actions › View Mode
code-oss adds the Agents window, a parallel multi-session interface for delegating, monitoring, and integrating agent tasks across projects︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ HOW TO FIND IT
Run two independent agent tasks in parallel — one implementing a feature, one investigating a bug — without waiting for either to finish before starting the other.
📍 1. Open the Agents window from the VS Code title bar.
2. Select New Session , choose This Computer , pick your first project folder, choose Worktree isolation from the isolation dropdown, and submit your first prompt.
3. Select New Session again (or hold it to keep both panes visible), choose your second project folder, and submit the second prompt.
4. Right-click either session in the sessions list and pin it so both panes remain visible side by side.
Review only the edits made in the current agent turn before committing, to catch scope creep before it lands in Git.
📍 1. Select the session in the sessions list to make it active.
2. Open the Secondary Side Bar and go to the changes view.
3. In the changeset dropdown, choose Last Turn to see only the most recent edits.
4. Confirm no unrelated files were changed, then switch to Uncommitted and select Commit .
code-oss adds the Agents window, a dedicated interface for running multiple agent sessions in parallel across projects, monitoring progress, and reviewing results in one place.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Run agent sessions with a choice of file isolation: Branch mode edits the current folder directly, while Worktree mode keeps changes in a separate Git worktree for independent parallel tasks. Not checked yet Arrange multiple session panes side by side by right-clicking a session and pinning it, or hold the New Session button to open an additional session without replacing the current pane. Not checked yet Review agent edits in the Secondary Side Bar using changeset filter options Branch , Uncommitted , Session , or Last Turn to scope the diff view to the relevant changes. Not checked yet Commit agent changes with the Commit action, which uses git commit --no-verify and skips pre-commitpre-commit An open-source framework for managing and running git hooks, maintained by the pre-commit project, that lets cyber tools enforce checks automatically on every commit before code reaches a repository. and commit-msg hooks; a GitHub Copilot sign-in is required to generate the commit message. Not checked yet
+2 more
show less
Continue a project session started in the Agents window inside the main editor window alongside the editor, debugger, and other development tools. Not checked yet Quick chats in the Agents window support tasks without a project workspace and can be continued in a local folder, with conversation history preserved, for Copilot and Codex harness sessions. Not checked yet
7 details in Everything
GitHub Copilot inline suggestions now support configurable eagerness: low, medium, and high modes controlling suggestion frequency and edit size.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
VS Code gains an Agent Customizations editor, Customization Marketplace, monorepo parent-repo discovery, and a quality-evaluation extension for customization files.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ USE IT
Enable monorepo parent-repo customization discovery so that shared copilot-instructions.md and skill files at the repo root are picked up when you open only a subfolder.
.vscode/settings.json
copy
chat.useCustomizationsInParentRepositories: true
Turn on the Customization Marketplace and the GitHub Feed source to browse and install community skills and MCP MCP Model Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. servers from within VS Code.
.vscode/settings.json
copy
chat.customizations.marketplace.enabled: true
chat.customizations.marketplace.sources.publicFeed.enabled: true
Analyze a customization file for contradictions and ambiguous wording before relying on it in production workflows.
📍 1. Open the target file (e.g. .github/instructions/style.instructions.md) in the editor.
2. Open the Command Palette (Ctrl+Shift+P / Cmd+Shift+P).
3. Run Chat Customizations Evaluations: Analyze .
4. Review diagnostics in the Problems panel (Ctrl+Shift+M / Cmd+Shift+M).
5. Select Implement Suggestions to apply recommended fixes.
Adds chat.customizations.marketplace.enabled setting to turn on a unified Discover page for finding skills, MCP servers, plugins, and other agent customizations from configured sources.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Adds chat.customizations.marketplace.sources.publicFeed.enabled setting to show or hide the GitHub Feed source in the Customization Marketplace. Not checked yet Adds chat.useCustomizationsInParentRepositories setting to discover customization files from parent repository folders (up to the .git root) when only a monorepo subfolder is opened — applies to copilot-instructions.md, AGENTS.md, CLAUDE.md, *.instructions.md, *.prompt.md, *.agent.md, and hooks. Not checked yet Adds the Agent Customizations editor, opened via the Chat: Open Customizations Command Palette command or the Customizations panel in the sidebar, providing a central place to create, manage, and scope customizations across User, Workspace, and Organization levels. Not checked yet Adds Chat: New <customization-type> Command Palette commands to create customization files directly in a supported location for each type. Not checked yet
+3 more
show less
Adds Developer: Open Agent Debug Logs Command Palette command (also reachable via the Chat view ellipsis menu > Show Agent Debug Logs ) to troubleshoot agent customization issues. Not checked yet Adds the Chat Customizations Evaluations extension (installed separately from the Visual Studio Marketplace) with a Chat Customizations Evaluations: Analyze command that uses GitHub Copilot to detect contradictions, ambiguous wording, persona conflicts, and intent gaps in SKILL.md, *.agent.md, *.instructions.md, and *.prompt.md files, surfacing diagnostics in the Problems panel. Not checked yet Supports AI-assisted customization creation: enter a natural-language prompt in the Agent Customizations editor Overview section to generate a new customization file, then review and save it. Not checked yet
8 details in Everything
Copilot harness sessions gain OS-level agent sandboxing sandboxing Running code inside a restricted environment that limits what it can access or change. The isolation reduces the impact of mistakes or untrusted code, depending on the sandbox's controls. , worktree isolation, and Autopilot permission level in code-oss︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ HOW TO FIND IT
Enable OS-level sandboxing for MCP MCP Model Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. servers and language servers in a Copilot harness session to confine their process execution.
📍 In a Copilot harness session, open Settings and confirm that both MCP server sandboxing and language server sandboxing settings are active (they are on by default). For remote sessions, verify prerequisites on the remote machine where commands execute.
Use Autopilot mode to let the agent auto-approve all tool calls and run unattended until the task is complete.
📍 In a Copilot harness session, open the tool-call approval picker and select Autopilot to enable fully autonomous task completion.
Copilot harness sessions support configurable tool-call approval levels, including manual approval, auto-approve, and an Autopilot mode that auto-approves all tools and drives the agent until the task is complete.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Copilot harness sessions using default tools gain OS-level agent sandboxing that confines shell execution, child processes, MCP servers, and language servers — available on Windows, macOS, and Linux, independent of permission level. Not checked yet Review pending agent edits in a diff editor before committing, merging, or creating a pull request; Local harness sessions and existing extension-host sessions also support keeping or undoing pending edits. Not checked yet
3 details in Everything
VS Code Copilot agent sessions gain per-session sandbox toggle, Claude permission modes, and Codex approval presets in the permissions picker.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ USE IT
Allow a Claude agent session to apply changes without any permission prompts inside a fully isolated, air-gapped dev container.
.vscode/settings.json
copy
"github.copilot.chat.claudeAgent.allowDangerouslySkipPermissions": true
Enable the sandbox for all new Copilot harness sessions on this machine so the agent cannot freely access the file system or network by default.
.vscode/settings.json
copy
"chat.agent.sandbox.enabled": true
Switch a Claude agent session to Plan mode so the agent outlines its approach before making any changes — useful for reviewing intent before execution.
📍 In the Copilot Chat panel, open the permissions picker and select Plan under Claude permission modes before starting the agent session.
Adds github.copilot.chat.claudeAgent.allowDangerouslySkipPermissions setting to bypass all Claude permission checks — recommended only in an isolated, air-gapped sandbox.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Exposes Claude harness permission modes (Edit automatically , Ask , Plan ) in the permissions picker for Claude agent sessions. Not checked yet Adds Codex approval presets (Default Permissions , Auto-Review , Full Access ) to the direct Codex integration via the permissions picker. Not checked yet Adds per-session sandbox toggle in the permissions picker for Copilot harness sessions, controlling sandboxing sandboxing Running code inside a restricted environment that limits what it can access or change. The isolation reduces the impact of mistakes or untrusted code, depending on the sandbox's controls. on or off for that session; managed settings can enforce sandboxing and disable the toggle. Not checked yet Surfaces chat.agent.sandbox.enabled as the configuration setting backing the sandbox behavior for Copilot harness sessions. Not checked yet
5 details in Everything
VS Code Agents adds Rubber Duck second-opinion critic and per-harness review workflows for inspecting, reverting, and integrating agent edits.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ TRY IT
Ask Rubber Duck to spot edge cases the agent missed before you commit its code changes.
copy
$ /rubber-duck What edge cases are missing?
Adds /rubber-duck <question> chat command to invoke the Rubber Duck critic, a read-only complementary-model reviewer that flags blocking issues, non-blocking issues, and suggestions on the agent's plans, code, and tests — without editing files or running commands.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Adds per-harness change-review workflows: Copilot/Claude/Codex harness sessions expose diffs in the Chat view, Agents window, or Source Control; Local harness sessions mark edits as pending for individual keep/revert; Cloud sessions return changes as a GitHub pull request. Not checked yet Adds a pill above the chat input showing affected-file count and diff statistics; selecting it opens a multi-diff editor with all session changes. Not checked yet Enables adding review comments to an open GitHub pull request directly from the Agents window for sessions using the Copilot or Claude harness, or the experimental Codex integration. Not checked yet Adds an undo-scope table covering inline edits, request-level revert, checkpoint restore, Source Control discard, and external-service recovery — letting practitioners choose recovery at the right granularity. Not checked yet
+2 more
show less
Supports customizing the diff view so that selecting a file opens a focused single-file diff instead of the default multi-file diff editor. Not checked yet Adds ⌥⌘B (macOS) / Ctrl+Alt+B (Windows/Linux) keyboard shortcut to open the Agents side panel. Not checked yet
Adds /rubber-duck <question> chat command to invoke the Rubber Duck critic, a read-only complementary-model reviewer that flags blocking issues, non-blocking issues, and suggestions on the agent's plans, code, and tests — without editing files or running commands.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Adds per-harness change-review workflows: Copilot/Claude/Codex harness sessions expose diffs in the Chat view, Agents window, or Source Control; Local harness sessions mark edits as pending for individual keep/revert; Cloud sessions return changes as a GitHub pull request. Not checked yet Adds a pill above the chat input showing affected-file count and diff statistics; selecting it opens a multi-diff editor with all session changes. Not checked yet Enables adding review comments to an open GitHub pull request directly from the Agents window for sessions using the Copilot or Claude harness, or the experimental Codex integration. Not checked yet Adds an undo-scope table covering inline edits, request-level revert, checkpoint restore, Source Control discard, and external-service recovery — letting practitioners choose recovery at the right granularity. Not checked yet
+1 more
show less
Adds ⌥⌘B (macOS) / Ctrl+Alt+B (Windows/Linux) keyboard shortcut to open the Agents side panel. Not checked yet
1 more in Everything
VS Code agents gain remote-machine delegation, browser-based management, Dev Container Dev Container A Docker-based development environment defined in a JSON config file, standardized by Microsoft, that lets a project ship its exact toolchain so any contributor's editor or CI system gets an identical workspace. support, and GitHub Mobile remote control for agent sessions.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ TRY IT
Take control of a running Copilot harness agent session from GitHub Mobile while away from your desk, without starting a new session.
Adds experimental remote-machine task delegation in the Agents window, letting the agent target connected remote machines automatically by criteria (Windows, Linux, or macOS, minimum memory, logical CPU CPU The main general-purpose processor in a computer. It runs application instructions and coordinates work with memory and other hardware. count, optional model) without manually selecting a machine.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Adds /remote on command inside a Copilot harness session to remotely monitor and steer that session from GitHub.com or the GitHub Mobile app, with session history, tool activity, status, approvals, and questions synchronized. Not checked yet Adds /remote off command to stop sharing a remotely controlled Copilot harness session. Not checked yet Supports running agents inside a remote project's Dev Container over SSH, Tunnel, and WSL WSL Windows Subsystem for Linux, a Microsoft-built compatibility layer that lets a Linux environment run directly on Windows without a virtual machine, giving cyber tools access to Linux binaries, filesystems, and syscalls on a Windows host. connections in the desktop Agents window, starting in VS Code 1.139. Not checked yet Adds browser-based Agents window access via dev tunnels, enabling session management from any device — including mobile — without installing VS Code locally. Not checked yet
+1 more
show less
Adds a hosts bar in the browser-based Agents window showing machines with configured dev tunnels, their connection status, and connect/disconnect controls; offline sessions reconnect automatically when the machine comes back online. Not checked yet
6 details in Everything
VS Code gains a searchable, cross-device Copilot session history that syncs to GitHub and spans CLI, cloud agent, code review, and Copilot Desktop.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
Adds a searchable history of Copilot harness sessions, synced by default to your GitHub account for cross-device access.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Enables sharing of selected sessions and unified querying of work from VS Code, Copilot CLI, cloud agent, code review, and the GitHub Copilot Desktop app. Not checked yet
2 details in Everything
VS Code agent sessions gain chat-to-folder/worktree moves, side chats, Copilot CLI terminal integration, and cross-session coordination tools.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ HOW TO FIND IT
Isolate a new feature implementation from the current conversation by moving it into a dedicated Git worktree before writing any code.
📍 Continue this task in a new Git worktree of the current repository before implementing it.
Resume a Copilot harness session in a terminal tab when you need CLI access alongside the ongoing chat.
📍 In the Agents window, right-click the Copilot harness session in the sessions list and select Resume in Terminal .
Start a fresh Copilot CLI session from VS Code without leaving the editor.
📍 Open the Command Palette and run Chat: New Copilot CLI Session , or open the GitHub Copilot CLI terminal profile from the terminal dropdown.
Adds natural-language prompts to move the current chat to an existing local folder (Move this chat to <absolute folder path> and continue the current task there.) or a new Git worktree (Continue this task in a new Git worktree of the current repository before implementing it.) while preserving conversation history.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Adds side chats for Copilot and Claude harness sessions in the Agents window, letting practitioners work on related tasks in peer chats without interrupting an ongoing conversation. Not checked yet Adds Chat: New Copilot CLI Session command and a GitHub Copilot CLI terminal profile to start a Copilot CLI session from VS Code, with a Resume in Terminal right-click action to continue an existing Copilot harness session from the terminal. Not checked yet Adds automatic cleanup settings (disabled by default) to remove inactive Copilot, Claude, and experimental Codex integration sessions after their pull requests merge. Not checked yet Adds cross-session coordination via built-in session-management tools available to agents in Copilot and Claude harness sessions, and with the experimental Codex integration. Not checked yet
+3 more
show less
Introduces session forking, which branches off a conversation inheriting the original history; forked sessions are prefixed with 'Forked:' and open as independent sessions when the harness does not support multiple chats. Not checked yet Adds support for switching between the Chat view and the Agents window surfaces after a session has already started. Not checked yet Adds Resume in Terminal context-menu action on existing Copilot harness sessions in the sessions list to continue them from an integrated terminal. Not checked yet
8 details in Everything
└──▷ BREAKING ON UPGRADE
! If enterprise-managed sandbox or permission settings automatically move a Local session to the Copilot harness, the original Local session becomes a read-only archived record and cannot be continued while the policy applies.
VS Code adds experimental cloud automations that run Copilot agent tasks on GitHub even when VS Code and your computer are off.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ USE IT
Enable cloud automation controls so you can schedule unattended Copilot agent runs on GitHub from within VS Code.
json
copy
chat.automations.cloud.enabled: true
After enabling cloud automations, create one targeting a private GitHub repo by selecting Cloud as the session type — walk through the VS Code Automations panel.
📍 1. Open the Automations view in VS Code.
2. Select Create a cloud automation .
3. For the workspace, select a local clone of a GitHub.com repository, or choose Work in GitHub to pick a repo not open locally.
4. Set Cloud as the session type.
5. Write the task prompt — include scope, expected output, and safety constraints.
6. Select the model and review allowed tools; grant only what the unattended task needs.
7. Set the schedule (start with manual to review the first run before enabling recurrence).
8. Review the configuration and confirm to create the automation.
Enables chat.automations.cloud.enabled to surface cloud automation controls, letting scheduled Copilot agent tasks run remotely on GitHub without requiring VS Code or your computer to stay on.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Adds a Cloud session type when creating an automation, targeting an eligible private GitHub.com repository instead of the local working tree. Not checked yet Adds Refresh and Open on GitHub actions on automation cards so practitioners can pull live status (queued, running, completed, failed) and manage runs directly on GitHub. Not checked yet AI-assisted automation creation is not available for cloud automations, and automations using unsupported GitHub event triggers are read-only in VS Code. Not checked yet
4 details in Everything
└──▷ BREAKING ON UPGRADE
! Turning off chat.automations.cloud.enabled hides cloud automation controls in VS Code but does NOT stop schedules already active on GitHub — disable or delete those cloud automations on GitHub first.
code-oss agents gain shell-profile initialization for Copilot harness sessions and editable tool-call confirmation dialogs︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ USE IT
Keep agent shell commands running in your real dev environment by loading your shell profile before each Copilot harness command — catches aliases, PATH entries, and env vars that a bare shell would miss.
json
copy
// In VS Code settings (settings.json)
{
"chat.agentHost.shellTool.initScript.enabled": true
}
Auto-activate the correct Python virtualenv for the workspace whenever the agent runs a shell command, pairing shell-profile init with the Python Environments extension.
json
copy
// In VS Code settings (settings.json)
{
"chat.agentHost.shellTool.initScript.enabled": true,
"python-envs.terminal.autoActivationType": "shellStartup"
}
Enables chat.agentHost.shellTool.initScript.enabled to load ~/.bashrc (macOS/Linux) or PowerShell profiles (Windows) before each command in Copilot harness sessions using the SDK built-in shell tool.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Supports parameter editing directly inside Local harness and extension-contributed tool confirmation dialogs, letting users adjust inputs before approving a tool call. Not checked yet With python-envs.terminal.autoActivationType set to shellStartup, the selected workspace Python environment is automatically activated alongside shell initialization. Not checked yet
Enables chat.agentHost.shellTool.initScript.enabled to load ~/.bashrc (macOS/Linux) or PowerShell profiles (Windows) before each command in Copilot harness sessions using the SDK built-in shell tool.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet
2 more in Everything
VS Code Insiders adds /research <topic> in Copilot harness sessions to gather findings from codebases, GitHub repos, and the web before planning.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ HOW TO FIND IT
Before committing to an implementation approach, gather a cited Markdown research report across your codebase, related GitHub repos, and the web.
📍 /research <topic>
Adds /research <topic> command Preview in VS Code Insiders Copilot harness sessions to gather findings from your codebase, relevant GitHub repositories, and the web, returning a detailed Markdown report with citations.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Introduces a built-in Plan agent in the Local harness session with distinct controls separate from Copilot Plan mode. Not checked yet
2 details in Everything
VS Code adds multi-harness agent selection (Copilot, Claude, Codex, Cloud), canvas extensions, and cross-harness handoffs in the Agents window.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ HOW TO FIND IT
Enable canvases before starting a Copilot harness session so the agent can create and update interactive visual surfaces in your workspace.
📍 Open Settings (Ctrl+,), search for chat.canvases.enabled, and toggle it on. Then open the Chat view and start a new Copilot harness session.
Generate a custom canvas UI for your project — useful when you want an agent-managed control panel or diagram tied to your current workspace.
Hand off a Local harness session to the Cloud target when you want a provider-managed agent to finish the work and return a pull request against your GitHub repository.
📍 In the Agents window, with a Local harness session active, select Cloud from the harness selector. VS Code carries the conversation history to the Cloud session, which then works against the selected GitHub repository.
Adds canvas support across the product: enable interactive canvases shared with the agent via chat.canvases.enabled for Copilot harness sessions, generate a canvas extension in your workspace with the /create-canvas skill, open canvases from Add Tab (+) > Canvas in the Agents window editor area, and install chat plugins containing canvas extensions from the GitHub Copilot app for use in VS Code.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Introduces the Agents window for selecting a harness (Copilot, Claude, Codex, or Cloud) and a GitHub repository or local folder before a session starts. Not checked yet Supports harness handoffs so a running session can continue with a different harness or Cloud target while carrying conversation history and context. Not checked yet Adds the Cloud execution target, which runs a provider-managed agent against a GitHub repository and returns a pull request without using local editor context. Not checked yet Integrates the Anthropic Claude Agent SDK, supporting BYOK BYOK Bring Your Own Key — a cloud deployment model where the customer supplies and controls the encryption keys rather than the provider, so sensitive data stays under the customer's cryptographic authority. -backed and Copilot-routed Claude models, and an experimental Agent Host integration for OpenAI Codex (available in the Agents window without the OpenAI Codex extension), as selectable harnesses within VS Code. Not checked yet
+1 more
show less
Identifies the active harness via a prefix in the chat title, a watermark in the Agents window chat input, and a tooltip in the sessions list. Not checked yet
6 details in Everything
VS Code Agents window gains worktree isolation, cloud automations, canvas support, and BYOK BYOK Bring Your Own Key — a cloud deployment model where the customer supplies and controls the encryption keys rather than the provider, so sensitive data stays under the customer's cryptographic authority. model configuration.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ USE IT
Keep an agent's changes isolated from your active workspace by defaulting new sessions to worktree mode in VS Code Insiders.
json
copy
// In VS Code Insiders settings (settings.json)
{
"sessions.useWorktree": true
}
Make large Git-ignored folders (e.g. node_modules) available inside a new worktree via a symlink so the agent can build and test without re-installing.
json
copy
// In VS Code settings (settings.json)
{
"git.worktreeSymlinkFolders": ["node_modules"]
}
Enable cloud automations so recurring agent tasks continue running in GitHub even after VS Code is closed.
json
copy
// In VS Code settings (settings.json)
{
"chat.automations.cloud.enabled": true
}
Adds sessions.useWorktree setting to control whether worktree isolation is selected by default when creating a new workspace session in VS Code Insiders, along with git.worktreeIncludeFiles to copy Git-ignored files (such as .env files and installed dependencies) into new worktrees and git.worktreeSymlinkFolders Experimental to symlink large ignored folders into a worktree instead of copying them, introducing worktree isolation for Agents window workspace sessions so agents can work in a separate Git worktree to keep changes out of the active workspace — useful for parallel tasks.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Adds chat.automations.cloud.enabled setting to enable cloud automations that run recurring agent tasks on a schedule (hourly, daily, or weekly) remotely in GitHub, continuing even after VS Code is closed. Not checked yet Supports Canvases Preview as a Copilot canvas type, bring-your-own-key (BYOK) models via provider configuration, and using Claude and Codex without GitHub sign-in Experimental , all in the Agents window. Not checked yet Enables running multiple chat tabs and split groups within a single Copilot or Claude harness session. Not checked yet Adds a pull-request creation form within a session to review details and choose post-creation actions. Not checked yet
5 details in Everything
VS Code agent sessions gain multi-chat support, cross-interface continuity, /delegate handoff, and local session discovery from Copilot CLI, Claude Code, and Codex.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ HOW TO FIND IT
Hand off a running task to a cloud agent without leaving your current Copilot harness session.
📍 In the Copilot harness session chat, type /delegate and describe the well-scoped task to continue it with a cloud agent.
Hand off an ongoing Local harness task to Claude with its full toolset and permissions.
📍 In the Agents window, open a Local harness session, open the Session Target dropdown, select Claude, and confirm the handoff.
Pick up a Claude Code session started outside VS Code without losing its conversation history.
📍 In the Agents window, filter by External to find the discovered Claude Code session, then send a message from VS Code to take ownership of it.
Supports running multiple chats in one session against the same codebase simultaneously, available with Copilot and Claude harnesses.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Enables moving an existing chat to a different folder or Git worktree in supported single-chat sessions in the Agents window, preserving full conversation history. Not checked yet Access and continue the same live Copilot or Claude session across both the Chat view and the Agents window without forking the conversation, harness, workspace, or worktree. Not checked yet Configuring the experimental Codex integration allows Codex sessions to be used across both interfaces. Not checked yet Enter /delegate in a Copilot harness session to hand off a well-scoped task to a cloud agent, available with Copilot, Claude, or Codex via the experimental integration. Not checked yet
+3 more
show less
Use the Session Target dropdown to hand off an ongoing task from a Local harness session to Copilot, Claude, Codex, or Cloud, with the destination using the selected harness's tools, permissions, and models. Not checked yet VS Code now discovers supported local sessions created in Copilot CLI, the GitHub Copilot app, Claude Code, and Codex; discovered sessions appear under the External filter until a message is sent from VS Code. Not checked yet Enables remote control for Copilot harness sessions, allowing users to monitor and steer a running session from GitHub. Not checked yet
8 details in Everything
VS Code 1.142 adds interactive canvases, portable chat context, expanded Markdown editor, and granular enterprise controls.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
Adds visual and interactive canvases for working with content in VS Code.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Enables moving chats without losing context. Not checked yet Expands the Markdown editor with new capabilities. Not checked yet Adds more granular enterprise controls. Not checked yet
Adds visual and interactive canvases for working with content in VS Code.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet
3 more in Everything
advisory CVE-2026-78461
Oct 13, 2026
·
issue 128
CVE-2026-78461: Security Feature Bypass via Notebooks (CVSS CVSS Common Vulnerability Scoring System, a method for describing a vulnerability's characteristics and severity. Its score is not the probability that attackers will exploit it. 7.4, affects < 1.136.2)︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet
advisory CVE-2026-69370
Oct 13, 2026
·
issue 128
CVE-2026-69370: Remote Code Execution via Workspace-injected Error Notification Link (CVSS CVSS Common Vulnerability Scoring System, a method for describing a vulnerability's characteristics and severity. Its score is not the probability that attackers will exploit it. 8.8, affects < 1.141.1)︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet
advisory CVE-2026-102229
Oct 13, 2026
·
issue 128
CVE-2026-102229: Information Disclosure via Remote Webview Resource Path Traversal (CVSS CVSS Common Vulnerability Scoring System, a method for describing a vulnerability's characteristics and severity. Its score is not the probability that attackers will exploit it. 6.8, affects < 1.141.1)︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet
advisory CVE-2026-102233
Oct 13, 2026
·
issue 128
CVE-2026-102233: Workspace Trust Bypass via Terminal Relaunch and Reuse (CVSS CVSS Common Vulnerability Scoring System, a method for describing a vulnerability's characteristics and severity. Its score is not the probability that attackers will exploit it. 7.8, affects < 1.141.1)︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet
VS Code adds agent sandboxing sandboxing Running code inside a restricted environment that limits what it can access or change. The isolation reduces the impact of mistakes or untrusted code, depending on the sandbox's controls. for Local and Copilot Agent Host sessions, with per-session toggles, network restrictions, and org-managed policy controls.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ HOW TO FIND IT
Turn on agent sandboxing for all Local and Copilot Agent Host sessions on your machine via the Settings editor.
📍 1. Open the Settings editor (Ctrl+, / Cmd+,).
2. Search for the sandbox enablement setting and set it to On.
3. Start a new session — existing Agent Host sessions require a toggle in the permissions picker to pick up the change.
Verify the sandbox policy actually in effect for the current Copilot Agent Host session after enabling or changing settings.
📍 1. In your active Copilot Agent Host session, run the 'inspect the effective sandbox policy' command.
2. Review the policy snapshot to confirm sandboxing is on and check which file system and network restrictions apply.
3. After any setting change, run the command again — each invocation produces a fresh snapshot; earlier links continue to show their original state.
Handle a blocked command by granting a session-scoped bypass rather than disabling sandboxing globally.
📍 1. When a terminal command is blocked, review the file system or network restriction shown in the notification.
2. If bypass is permitted by both managed policy and your local settings, approve the agent's request on the Allow in This Session prompt.
3. Sandboxing is suspended for subsequent terminal commands in this session only — other sessions and User/Workspace settings are not changed.
4. To re-enable sandboxing for the session, use the permissions picker toggle; a later bypass will require a new approval.
Adds sandboxing support for Local sessions and Copilot Agent Host sessions, controlling file system and network access for terminal commands run by agents.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Introduces a permissions picker toggle to change the current Agent Host session's sandbox state without affecting other sessions or global settings. Not checked yet Supports per-session sandbox state that persists across session restores, window reloads, and VS Code restarts unless overridden by a managed policy. Not checked yet Adds an Allow in This Session bypass prompt that turns off sandboxing session-wide only after explicit approval, leaving other sessions and global settings unchanged. Not checked yet Adds organization-managed sandboxing controls that can require sandboxing, prevent bypass, or block outbound network access — enforced settings are locked with an org-managed indicator. Not checked yet
+2 more
show less
Adds network restriction behavior that varies by environment: the Copilot Agent Host built-in shell and Windows terminal sandbox block outbound access entirely, while Local sessions and the Agent Host custom terminal tool on macOS, Linux, or WSL2 WSL2 A Microsoft-maintained compatibility layer in Windows 10/11 that runs a real Linux kernel inside a lightweight VM, letting cyber tools built for Linux execute natively on Windows without dual-booting or a full VM. use domain allowlist filtering where an empty allowlist blocks all outbound access. Not checked yet Supports a inspect the effective sandbox policy command on Copilot Agent Host sessions to snapshot the current policy and restrictions; re-running produces a new snapshot while earlier links preserve their original state. Not checked yet
7 details in Everything
VS Code gains remote agent session delegation and chat-without-workspace on remote hosts, both experimental.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ USE IT
Enable remote agent delegation so an agent can automatically place work on the least-loaded connected host.
json
copy
// In VS Code settings (settings.json)
{
"chat.remoteAgentHostsEnabled": true,
"chat.remoteSessions.tools.enabled": true
}
Enable workspace-less chat on a remote host when you need its environment but have no repository to open.
json
copy
// In VS Code settings (settings.json)
{
"sessions.chat.unifiedWorkspacePicker.enabled": true
}
Start a delegated task on a remote Linux host with at least 16 GB of memory, then retrieve its result.
📍 1. Open the Agents window and start an Agent Host session.
2. In your prompt, instruct the agent to use list_agent_hosts to find available remote hosts.
3. Ask the agent to use create_remote_session with criteria: Linux, minimum memory 16 GB.
4. After work completes, ask the agent to use get_remote_session to retrieve the session status and final response.
5. To receive results in the originating chat, instruct the remote agent to call send_remote_message with its result.
Enables sessions.chat.unifiedWorkspacePicker.enabled to start a chat without a workspace on a remote SSH or dev-tunnel host via the workspace picker Chat entry.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Adds chat.remoteAgentHostsEnabled and chat.remoteSessions.tools.enabled settings to let an agent delegate work to connected remote hosts from an Agent Host session. Not checked yet Adds list_agent_hosts tool so an agent can discover connected remote hosts, available models, resource capacity, and session load. Not checked yet Adds create_remote_session tool to start a session on a specific remote host or auto-place it by criteria (Windows, Linux, or macOS, minimum memory, logical CPU CPU The main general-purpose processor in a computer. It runs application instructions and coordinates work with memory and other hardware. count, optional model), favoring the host with the fewest running sessions. Not checked yet Adds get_remote_session tool to check a remote session's status and latest response. Not checked yet
+2 more
show less
Adds send_remote_message tool to send follow-up work or report results and questions back to the exact originating chat, with queue support if that chat is busy. Not checked yet Turning off chat.remoteSessions.tools.enabled removes the remote delegation tools without disconnecting hosts or stopping already-running remote sessions. Not checked yet
Enables sessions.chat.unifiedWorkspacePicker.enabled to start a chat without a workspace on a remote SSH or dev-tunnel host via the workspace picker Chat entry.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Adds chat.remoteAgentHostsEnabled and chat.remoteSessions.tools.enabled settings to let an agent delegate work to connected remote hosts from an Agent Host session. Not checked yet Adds list_agent_hosts tool so an agent can discover connected remote hosts, available models, resource capacity, and session load. Not checked yet Adds create_remote_session tool to start a session on a specific remote host or auto-place it by criteria (Windows, Linux, or macOS, minimum memory, logical CPU CPU The main general-purpose processor in a computer. It runs application instructions and coordinates work with memory and other hardware. count, optional model), favoring the host with the fewest running sessions. Not checked yet Adds get_remote_session tool to check a remote session's status and latest response. Not checked yet
+1 more
show less
Adds send_remote_message tool to send follow-up work or report results and questions back to the exact originating chat, with queue support if that chat is busy. Not checked yet
1 more in Everything
VS Code agent sessions gain peer chats, per-chat done state, context compaction context compaction Condensing an AI conversation or working history to free space in the model's context window. A summary preserves useful information, but details can be lost. , and orchestration-limit controls︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ HOW TO FIND IT
Lift all agent orchestration limits so a coordination-heavy multi-agent workflow can spawn sessions and send messages without hitting defaults.
📍 Set chat.agentHost.agentOrchestrationLimits to off in VS Code settings (no Agent Host restart required).
Show every peer chat as its own tab so you can monitor two concurrent agent conversations side by side in split groups.
📍 Set sessions.showChatTabs to Multiple in VS Code settings, then use the tab strip's context menu to drag chat tabs into a split group.
Ask an agent to spin up an isolated peer chat for read-only research without touching the files in your current checkout.
📍 Create a peer chat in this session that reuses the current workspace and checkout to review the authentication flow. Do not modify files.
Adds sessions.showChatTabs config key to control how peer chats appear — Multiple (each chat as a tab, arrangeable in split groups) or Single (active chat fills the pane).︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Adds chat.agentHost.agentOrchestrationLimits config key to control process-wide limits on sessions, chats, messages, and recursive session creation; supports on (default) and off values, effective without restarting the Agent Host. Not checked yet Adds peer chats to Agent Host sessions — create additional chats that share the session's workspace and worktree without interrupting the main conversation, via right-click > New Chat in This Session on an active session. Not checked yet Adds per-chat mark as done action for peer chats inside an Agent Host session, hiding the chat from the active list while preserving its title and transcript for later restoration. Not checked yet Adds context compaction, which summarizes earlier conversation history to free context-window space and reduce tokens sent with subsequent requests. Not checked yet
+5 more
show less
Adds a context window context window The maximum amount of text, counted in tokens, a model can take in and produce in one request. Everything the model knows about the task has to fit inside it. control in the chat input showing token count, usage breakdown by category, and total AI credits consumed by the session. Not checked yet Adds currentSession and independent session-relationship options for agent-created sessions, letting prompts explicitly request a peer chat in the current session or a fully separate session. Not checked yet Adds external Codex session conflict detection — when a chat is simultaneously open in ChatGPT or Codex CLI, VS Code shows a 'This chat is open in another app' banner and disables sending until the other app is quit and Retry is selected. Not checked yet Adds Open as Editor and Open to the Side context-menu actions on peer-chat rows in the Sessions view, enabling peer chats to be opened in editor tabs or split editor groups. Not checked yet Adds Show Done Chats context-menu action on a session in the sessions list to surface completed peer chats beneath their owning session for restoration. Not checked yet
10 details in Everything
└──▷ USE IT
Preserve a tool's original column-width output in chat when long lines are getting wrapped and cutting off content.
json
copy
"chat.tools.terminal.outputReflow": false
└──▷ USE IT
Enable persistent progress with verbose tool-call details to monitor every step of a long-running agent session.
json
copy
chat.experimental.persistentProgress: Draw
chat.experimental.persistentProgressVerbosity: Verbose
VS Code Agents window gains experimental welcome phrases and grouped composer controls with new config keys and keyboard shortcuts︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ USE IT
Enable the grouped composer layout so workspace, branch, worktree, and harness controls appear above the chat input in the Agents window.
json
copy
{
"sessions.chat.unifiedWorkspacePicker.enabled": true,
"sessions.chat.experimental.newSessionComposerLayout": true
}
Show a rotating welcome phrase and personalize it with your name in the Agents window new-session composer.
json
copy
{
"sessions.chat.experimental.welcomePhrases": true,
"sessions.chat.experimental.welcomeName": "Alex"
}
Suppress the screen-reader announcement of the welcome heading when it appears in the Agents window.
json
copy
{
"accessibility.verbosity.newSessionWelcome": false
}
Adds sessions.chat.experimental.welcomePhrases to show a rotating welcome phrase above the new-session composer in the Agents window.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Adds sessions.chat.experimental.welcomeName (also settable via the Set Welcome Name action) to personalize the welcome heading; the name syncs across devices. Not checked yet Adds accessibility.verbosity.newSessionWelcome to control whether the welcome heading is announced in screen reader optimized mode. Not checked yet Adds sessions.chat.experimental.newSessionComposerLayout to group workspace, branch, worktree, and harness controls above the new-session chat input (requires sessions.chat.unifiedWorkspacePicker.enabled). Not checked yet Adds ⌘K ⌘F / Ctrl+K Ctrl+F to focus the workspace picker and ⌘K ⌘H / Ctrl+K Ctrl+H to focus the harness picker in the Agents window. Not checked yet
5 details in Everything
code-oss: Adds chat.agentHost.devContainer.idleTimeout config setting to control container idle shutdown period; set to 0 to disable automatic shutdown entirely (default is 300 seconds).︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ USE IT
Prevent dev containers from shutting down mid-session when long-running checks keep a PR open but no active turns are in progress.
yaml
copy
chat.agentHost.devContainer.idleTimeout: 0
Enable agent merge on an existing PR so VS Code automatically asks the agent to fix blockers and merges when checks pass.
📍 In the Agents window, open the session associated with your pull request › select Agent merge › Enable agent merge › review options in Configure agent merge for Active Session › confirm.
Adds an Agent merge Changes changeset dropdown entry that compares the latest completed agent merge repair turn against the preceding completed user turn, making repair diffs reviewable in one click. Not checked yet
1 detail in Everything
VS Code Agent Host chat now supports peer chats within a session and per-session target, agent, model, and permissions controls.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ HOW TO FIND IT
Start a coding task in Agent Host with safe, manual control over edits — choose Copilot as target, Auto as agent, and Manual permissions before sending the first prompt.
📍 In the Chat view input area, set Session Target to Copilot , Agent to Auto , Language model to your preferred model, and Permissions to Manual permissions , then type your first prompt.
Switch between peer chats in a running session without losing context or starting a new session.
📍 In the Sessions view, expand the main chat row to reveal peer chats, then select a peer chat to open its conversation in the Chat view.
Enables the Copilot session target to apply and save edits directly to a selected folder or worktree, with per-request changed-files summaries available. Not checked yet
1 detail in Everything
code-oss: A separate network domain filtering setting applies to the fetch tool and integrated browser independently of terminal sandboxing sandboxing Running code inside a restricted environment that limits what it can access or change. The isolation reduces the impact of mistakes or untrusted code, depending on the sandbox's controls. .
VS Code adds chat.tools.terminal.outputReflow to control whether terminal output previews reflow to fit chat width.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ USE IT
Preserve the terminal's original column width in chat output previews when long command output is being truncated or reflowed undesirably.
settings.json
copy
// In VS Code settings (settings.json)
"chat.tools.terminal.outputReflow": false
code-oss adds experimental persistent progress indicator for agent responses with new config keys︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ USE IT
Enable the persistent progress indicator in Stable builds and keep full tool-call details visible throughout an agent response.
json
copy
"chat.experimental.persistentProgress": "Draw",
"chat.experimental.persistentProgressVerbosity": "Verbose"
Adds chat.experimental.persistentProgress config key to keep a progress indicator visible until an agent response finishes, with Draw (colored or monochrome animation) or Off options.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Adds chat.experimental.persistentProgressVerbosity config key to control completed tool-call detail visibility: Compact replaces completed groups with expandable summaries, Verbose keeps full details visible. Not checked yet With persistent progress enabled, reasoning appears in separate collapsible previews and tool calls remain visible while running. Not checked yet Adds Open Accessible View command (⌥F2 / Alt+F2 / Shift+Alt+F2 ) to inspect a chat response character by character or line by line with a keyboard or screen reader. Not checked yet
Adds chat.experimental.persistentProgress config key to keep a progress indicator visible until an agent response finishes, with Draw (colored or monochrome animation) or Off options.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Adds chat.experimental.persistentProgressVerbosity config key to control completed tool-call detail visibility: Compact replaces completed groups with expandable summaries, Verbose keeps full details visible. Not checked yet Adds Open Accessible View command (⌥F2 / Alt+F2 / Shift+Alt+F2 ) to inspect a chat response character by character or line by line with a keyboard or screen reader. Not checked yet
1 more in Everything
VS Code Agents window adds agent merge for automated PR unblocking and multi-chat split groups in Agent Host sessions.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ USE IT
Prevent idle agent containers from being automatically shut down when long-running checks or approval prompts are expected.
yaml
copy
chat.agentHost.devContainer.idleTimeout: 0
Enable agent merge on an existing PR so the agent continuously addresses blockers until the PR is ready to merge.
📍 In the Agents window, open the session for the PR, select Agent merge › Enable agent merge , then review options under Configure agent merge for Active Session before confirming.
Inspect what the agent changed in its latest PR repair cycle before approving a merge.
📍 In the Agents window, open the changeset dropdown for the session and select Agent merge Changes to compare the latest agent merge repair turn with the preceding user turn.
Adds the Agents window, a dedicated VS Code window for tracking and reviewing agent sessions across multiple projects without switching between project windows.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Adds agent merge, an experimental feature that monitors a PR associated with an agent session and automatically asks the agent to address blockers until the PR is ready to merge. Not checked yet Adds the Agent merge Changes changeset in the changeset dropdown to review changes from the most recent agent merge repair cycle. Not checked yet Adds support for chat tabs and split groups in Agent Host sessions, allowing multiple conversations (peer chats, side chats, read-only subagent chats) to be arranged horizontally or vertically. Not checked yet
4 details in Everything
VS Code Agent chat view gains peer chat switching, Session Target controls, and a changed-files summary toggle.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ HOW TO FIND IT
Switch to a peer chat inside an active Agent Host session without losing context or starting over.
📍 In the Chat view, go to the Sessions view, expand the main chat row, and select a peer chat entry to open its conversation.
Configure a first coding task with recommended defaults before sending any prompt.
📍 In the chat input area, set Session Target to Copilot , Agent to Auto , Language model to your preferred model, and Permissions to Manual permissions , then type your prompt.
Adds peer chat switching in the Sessions view: expand a main chat row to open a peer conversation in the Chat view without starting a new session.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Adds Session Target , Agent , Language model , and Permissions controls to the chat input area for configuring each session before the first prompt. Not checked yet Adds a changed-files summary toggle to the Copilot session target, displaying a summary of changed files after each request. Not checked yet Supports Anthropic Claude and OpenAI Codex model credentials via agent harness agent harness The software around an AI agent that manages its tools, execution loop, state and limits. It determines how the model's proposed actions are run and checked. authentication options for supported workflows. Not checked yet
4 details in Everything
VS Code gains a Migrate MCP MCP Model Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. Servers workflow to move MCP server configs to portable .mcp.json or $COPILOT_HOME/mcp-config.json files.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ HOW TO FIND IT
Migrate MCP servers from VS Code-specific config to a portable workspace file so the configuration can be shared across agent hosts.
📍 1. Open the Agent Customizations editor.
2. Select an Agent Host — a migration card appears if VS Code detects servers to migrate.
3. Select Migrate MCP Servers .
4. Review the Migratable and Not migratable groups; for any partially supported server select Edit Configuration to resolve compatibility issues (e.g. explicit working directory, SSE transport).
5. Confirm migration — compatible servers are written to the workspace .mcp.json file with their enabled/disabled state preserved.
6. Review the destination .mcp.json before removing the original configuration, as the files do not stay synchronized.
Adds Migrate MCP Servers action in the Agent Customizations editor to move MCP servers to a portable workspace .mcp.json file or user $COPILOT_HOME/mcp-config.json file, preserving each server's enabled/disabled state.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Pre-migration compatibility check separates servers into Migratable and Not migratable groups, flagging servers with unsupported configuration such as explicit working directories, SSE SSE Server-Sent Events, a browser API and HTTP-based protocol where a server pushes a stream of text events to a client over a single long-lived connection, giving tools a lightweight alternative to WebSockets for one-way real-time updates. transport, or server version metadata. Not checked yet
2 details in Everything
VS Code agent sandboxing sandboxing Running code inside a restricted environment that limits what it can access or change. The isolation reduces the impact of mistakes or untrusted code, depending on the sandbox's controls. arrives for Local and Copilot Agent Host sessions, with per-session toggles, network controls, and org-managed policy enforcement.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ HOW TO FIND IT
Turn on agent sandboxing for a new Local or Copilot Agent Host session so terminal commands run inside a file-system- and network-restricted sandbox.
📍 1. Open the Settings editor (Ctrl+, / Cmd+,).
2. Search for the agent sandbox enablement setting and set it to on.
3. Start a new agent session — sandboxing is now active. For an existing Agent Host session, use the session toggle in the permissions picker to enable it for that session.
Verify the effective sandbox policy for an active Copilot Agent Host session after changing a setting, to confirm restrictions are applied as expected.
📍 1. In an active Copilot Agent Host session, run the sandbox policy inspection command from the command palette.
2. Review the policy snapshot — each report is a point-in-time view; run the command again after any setting change to see the updated policy.
Adds agent sandboxing for Local sessions and Copilot Agent Host sessions, controllable via the Settings editor with a dedicated enablement setting that defaults to off.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Adds a session toggle in the permissions picker for Copilot Agent Host sessions to change the current session's sandbox state without altering User or Workspace settings. Not checked yet Supports per-session sandbox policy inspection for Copilot Agent Host sessions via a dedicated command that returns a snapshot of the effective sandbox policy, including active restrictions. Not checked yet Supports a session-wide sandbox bypass approval flow: Allow in this Session disables sandboxing for subsequent terminal commands in that session after approval, without affecting other sessions or global settings. Not checked yet Outbound network behavior varies by environment: the Copilot Agent Host built-in shell and Windows terminal sandbox block outbound access unconditionally, while Local sessions and the Agent Host custom terminal tool on macOS, Linux, or WSL2 WSL2 A Microsoft-maintained compatibility layer in Windows 10/11 that runs a real Linux kernel inside a lightweight VM, letting cyber tools built for Linux execute natively on Windows without dual-booting or a full VM. use domain filtering with an empty allowlist blocking all outbound access. Not checked yet
5 details in Everything
code-oss gains remote agent delegation tools and workspace-free remote chat sessions with SSH/dev-tunnel support︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ HOW TO FIND IT
Start a chat on a remote host when you need its environment but have no repo to open — useful for running scripts or ad-hoc tasks on a beefy build server.
📍 1. Open the workspace picker in code-oss.
2. Enable sessions.chat.unifiedWorkspacePicker.enabled in Settings.
3. Connect the remote host over SSH or a dev tunnel.
4. In the workspace picker, select the Chat entry for the connected host (do NOT select a folder).
5. Choose an agent harness, enter your prompt, and press Enter.
Enable experimental remote agent delegation so an orchestrating agent can automatically farm sub-tasks out to the best-fit connected host.
json
copy
{
"chat.remoteAgentHostsEnabled": true,
"chat.remoteSessions.tools.enabled": true
}
Instruct an agent to auto-place a delegated session on a Linux host with at least 16 GB RAM RAM Working memory that holds programs and data while a computer is running. Its contents usually disappear when power is removed. , then poll for its result and relay it back.
📍 1. From an Agent Host session in the Agents window, prompt the agent:
'Use create_remote_session targeting Linux with minimum 16 GB memory to run the analysis.'
2. After the remote session completes, prompt:
'Use send_remote_message to forward the result back to this chat.'
3. Monitor progress with get_remote_session as needed.
Enables chat sessions on a remote host without a workspace by setting sessions.chat.unifiedWorkspacePicker.enabled and connecting over SSH or a dev tunnel, then selecting the Chat entry for that host in the workspace picker.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Adds experimental remote agent delegation via chat.remoteAgentHostsEnabled and chat.remoteSessions.tools.enabled, letting an Agent Host session delegate tasks to connected remote hosts without manually selecting the host in the workspace picker. Not checked yet Adds list_agent_hosts tool so an agent can discover connected remote hosts, available models, resource capacity, and current session load. Not checked yet Adds create_remote_session tool to start a session on a specific remote host or auto-place it by criteria — OS (Windows, Linux, macOS), minimum memory, logical CPU CPU The main general-purpose processor in a computer. It runs application instructions and coordinates work with memory and other hardware. count, and optional model — favoring the host with the fewest running and pending sessions. Not checked yet Adds get_remote_session tool to check a remote session's status and latest response. Not checked yet
+2 more
show less
Adds send_remote_message tool to send follow-up work or report results and questions back to the exact originating chat; if that chat is busy, the message queues. Not checked yet Disabling chat.remoteSessions.tools.enabled removes the remote delegation tools without disconnecting hosts or stopping already-running remote sessions. Not checked yet
Enables chat sessions on a remote host without a workspace by setting sessions.chat.unifiedWorkspacePicker.enabled and connecting over SSH or a dev tunnel, then selecting the Chat entry for that host in the workspace picker.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Adds experimental remote agent delegation via chat.remoteAgentHostsEnabled and chat.remoteSessions.tools.enabled, letting an Agent Host session delegate tasks to connected remote hosts without manually selecting the host in the workspace picker. Not checked yet Adds list_agent_hosts tool so an agent can discover connected remote hosts, available models, resource capacity, and current session load. Not checked yet Adds create_remote_session tool to start a session on a specific remote host or auto-place it by criteria — OS (Windows, Linux, macOS), minimum memory, logical CPU CPU The main general-purpose processor in a computer. It runs application instructions and coordinates work with memory and other hardware. count, and optional model — favoring the host with the fewest running and pending sessions. Not checked yet Adds get_remote_session tool to check a remote session's status and latest response. Not checked yet
+1 more
show less
Adds send_remote_message tool to send follow-up work or report results and questions back to the exact originating chat; if that chat is busy, the message queues. Not checked yet
1 more in Everything
VS Code agents gain peer chats, context compaction context compaction Condensing an AI conversation or working history to free space in the model's context window. A summary preserves useful information, but details can be lost. , per-chat done state, and chat.agentHost.agentOrchestrationLimits orchestration controls.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ HOW TO FIND IT
Switch peer chats to tab layout so you can monitor multiple agent conversations side by side in the Agents window.
📍 In VS Code Settings, search for sessions.showChatTabs and set it to Multiple . In the Agents window, each open peer chat now appears as its own tab — drag tabs into split groups to view conversations side by side.
Remove agent orchestration limits to support coordination-heavy workflows where agents spawn many sessions or send high message volumes.
📍 In VS Code Settings, search for chat.agentHost.agentOrchestrationLimits and set it to off. Changes take effect without restarting the Agent Host.
Ask an agent to spin up a related peer chat that reads the current repo without touching files, keeping your main conversation uninterrupted.
📍 Create a peer chat in this session that reuses the current workspace and checkout to review the authentication flow. Do not modify files.
Adds sessions.showChatTabs config setting to control whether peer chats appear as tabs (Multiple ) or replace the active pane (Single ) in the Agents window.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Adds chat.agentHost.agentOrchestrationLimits setting to control process-wide limits for agent-created sessions and chats, messages sent, and recursive session creation — set to off to remove limits. Not checked yet Adds currentSession and independent prompt keywords so agents can create a peer chat in the current session or a fully separate session, with optional worktree isolation. Not checked yet Introduces peer chats within an Agent Host session, letting practitioners run multiple independent conversations that share the same workspace and worktree without interrupting an ongoing chat. Not checked yet Adds context compaction, which summarizes earlier conversation history to free context-window space and reduce tokens sent with subsequent requests. Not checked yet
+7 more
show less
Adds a context window context window The maximum amount of text, counted in tokens, a model can take in and produce in one request. Everything the model knows about the task has to fit inside it. control in the chat input showing token count, a usage breakdown by category, and total AI credits consumed by the session. Not checked yet Adds the ability to mark an individual peer chat as done independently of the main chat, hiding it from the active sessions list while preserving its title and transcript for later restoration. Not checked yet Adds New Chat in This Session right-click action on active sessions, plus a session-header menu entry and keyboard shortcut, to create a peer chat. Not checked yet Adds Show Done Chats right-click action on a session to surface done peer chats and restore them with full history intact. Not checked yet Adds external Codex session conflict detection: when a chat is open in ChatGPT or Codex CLI, VS Code shows a 'This chat is open in another app' banner and disables sending until the other app is quit. Not checked yet Supports opening a peer chat as an editor tab or in a split editor group via right-click Open as Editor and Open to the Side actions in the Sessions view. Not checked yet For Dev Container Dev Container A Docker-based development environment defined in a JSON config file, standardized by Microsoft, that lets a project ship its exact toolchain so any contributor's editor or CI system gets an identical workspace. sessions, marking a session as done now removes the container immediately (without waiting for the idle shutdown period) when no other session or unsent draft uses it. Not checked yet
Adds sessions.showChatTabs config setting to control whether peer chats appear as tabs (Multiple ) or replace the active pane (Single ) in the Agents window.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Adds chat.agentHost.agentOrchestrationLimits setting to control process-wide limits for agent-created sessions and chats, messages sent, and recursive session creation — set to off to remove limits. Not checked yet Adds currentSession and independent prompt keywords so agents can create a peer chat in the current session or a fully separate session, with optional worktree isolation. Not checked yet Introduces peer chats within an Agent Host session, letting practitioners run multiple independent conversations that share the same workspace and worktree without interrupting an ongoing chat. Not checked yet Adds context compaction, which summarizes earlier conversation history to free context-window space and reduce tokens sent with subsequent requests. Not checked yet
+6 more
show less
Adds a context window context window The maximum amount of text, counted in tokens, a model can take in and produce in one request. Everything the model knows about the task has to fit inside it. control in the chat input showing token count, a usage breakdown by category, and total AI credits consumed by the session. Not checked yet Adds the ability to mark an individual peer chat as done independently of the main chat, hiding it from the active sessions list while preserving its title and transcript for later restoration. Not checked yet Adds New Chat in This Session right-click action on active sessions, plus a session-header menu entry and keyboard shortcut, to create a peer chat. Not checked yet Adds Show Done Chats right-click action on a session to surface done peer chats and restore them with full history intact. Not checked yet Adds external Codex session conflict detection: when a chat is open in ChatGPT or Codex CLI, VS Code shows a 'This chat is open in another app' banner and disables sending until the other app is quit. Not checked yet Supports opening a peer chat as an editor tab or in a split editor group via right-click Open as Editor and Open to the Side actions in the Sessions view. Not checked yet
1 more in Everything
VS Code adds chat.tools.terminal.outputReflow to control whether terminal output previews reflow in chat.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ USE IT
Preserve exact terminal column width in chat output previews when working with wide tables or log lines that must not wrap.
json
copy
"chat.tools.terminal.outputReflow": false
Adds experimental persistent progress indicator for agent responses with two new config keys in VS Code chat︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ USE IT
Enable the persistent progress indicator with verbose tool-call details to monitor every step of a long-running agent response.
json
copy
"chat.experimental.persistentProgress": "Draw",
"chat.experimental.persistentProgressVerbosity": "Verbose"
Adds chat.experimental.persistentProgress config key to keep a progress indicator visible until an agent response finishes, with Draw (colored or monochrome animation) and Off options; defaults to Draw in Insiders and Off in Stable.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Adds chat.experimental.persistentProgressVerbosity config key to control completed tool-call detail visibility: Compact replaces completed groups with expandable summaries, Verbose keeps full details visible. Not checked yet With persistent progress enabled, reasoning appears in separate collapsible previews and tool calls remain visible while they run. Not checked yet
3 details in Everything
VS Code Agents window gains experimental welcome phrases, personalized names, and grouped composer controls︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ USE IT
Enable a rotating personalized welcome heading in the Agents window, using your name instead of a generic phrase.
json
copy
// In VS Code settings (JSON)
{
"sessions.chat.experimental.welcomePhrases": true,
"sessions.chat.experimental.welcomeName": "Alex"
}
Group all session-context controls above the chat input for a cleaner Agents window layout.
json
copy
// In VS Code settings (JSON)
{
"sessions.chat.unifiedWorkspacePicker.enabled": true,
"sessions.chat.experimental.newSessionComposerLayout": true
}
Silence the screen-reader announcement of the welcome heading if it becomes repetitive during a long session.
json
copy
// In VS Code settings (JSON)
{
"accessibility.verbosity.newSessionWelcome": false
}
Adds sessions.chat.experimental.welcomePhrases setting to show a rotating welcome phrase above the new-session composer in the Agents window.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Adds sessions.chat.experimental.welcomeName setting (also settable via the Set Welcome Name action) to personalize the welcome heading; syncs across devices and falls back to your GitHub profile first name. Not checked yet Adds accessibility.verbosity.newSessionWelcome setting to control whether the welcome heading is announced in screen-reader optimized mode. Not checked yet Adds sessions.chat.experimental.newSessionComposerLayout setting to group workspace, branch, worktree, and harness controls above the new-session chat input (requires sessions.chat.unifiedWorkspacePicker.enabled). Not checked yet Adds ⌘K ⌘F / Ctrl+K Ctrl+F keybinding to focus the workspace picker and ⌘K ⌘H / Ctrl+K Ctrl+H to focus the harness picker, assignable via Configure Keybinding in the picker's context menu. Not checked yet
5 details in Everything
VS Code Agents window gains Agent Merge (automated PR blocker resolution) and configurable idle container shutdown via chat.agentHost.devContainer.idleTimeout.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ USE IT
Prevent the agent container from shutting down automatically during long-running, unattended workflows.
.vscode/settings.json
copy
"chat.agentHost.devContainer.idleTimeout": 0
Enable Agent Merge to automatically resolve PR blockers and merge when all checks pass, without manual intervention.
📍 1. Open the Agents window and locate your active session.
2. Select Agent merge to open Configure agent merge for Active Session .
3. Review agent merge settings, then select Enable agent merge .
4. Monitor progress — Agent merge addresses blockers and merges the PR when ready.
Introduces Agent Merge, an experimental feature that monitors a pull request associated with an agent session and autonomously asks the agent to address blockers until the PR is ready to merge.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Adds an Agent merge Changes changeset in the changeset dropdown to review changes from the most recent agent merge repair cycle. Not checked yet Supports chat tabs and split groups (horizontal or vertical) within Agent Host sessions, enabling multiple simultaneous conversations including interactive peer chats, side chats, and read-only subagent chats. Not checked yet
3 details in Everything
Agent Host sessions now support peer chats alongside the main chat, switchable from the Sessions view.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ HOW TO FIND IT
Switch between the main chat and a peer chat within an active Agent Host session without losing context.
📍 In the Sessions view, expand the main chat row, then select a peer chat entry to open its conversation in the Chat view.
VS Code agent sandboxing sandboxing Running code inside a restricted environment that limits what it can access or change. The isolation reduces the impact of mistakes or untrusted code, depending on the sandbox's controls. adds OS-level file system and network isolation for terminal commands, with domain filtering on macOS and Linux.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
Introduces agent sandboxing that uses operating system-level isolation to restrict file system and network access for terminal commands and their child processes in Local and Copilot Agent Host sessions.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Local terminal sandbox is in Preview on macOS, Linux, and WSL2 WSL2 A Microsoft-maintained compatibility layer in Windows 10/11 that runs a real Linux kernel inside a lightweight VM, letting cyber tools built for Linux execute natively on Windows without dual-booting or a full VM. , and Experimental on Windows; sandboxing for the Copilot Agent Host built-in shell is Experimental on all supported platforms. Not checked yet A separate setting provides network domain filtering for the fetch tool and integrated browser, independently of terminal sandboxing. Not checked yet
3 details in Everything
code-oss: Enables chat.agentHost.devContainer.enabled to run agent sessions inside a remote project's Dev Container Dev Container A Docker-based development environment defined in a JSON config file, standardized by Microsoft, that lets a project ship its exact toolchain so any contributor's editor or CI system gets an identical workspace. on SSH, Tunnel, and WSL WSL Windows Subsystem for Linux, a Microsoft-built compatibility layer that lets a Linux environment run directly on Windows without a virtual machine, giving cyber tools access to Linux binaries, filesystems, and syscalls on a Windows host. hosts, available from VS Code 1.139.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ HOW TO FIND IT
Run an agent session inside a remote project's Dev Container so the agent uses the project's configured tools and dependencies without installing them on the host.
📍 1. Open VS Code Settings and set chat.agentHost.devContainer.enabled to true.
2. Connect to an SSH, Tunnel, or WSL host and select the project folder in the workspace picker.
3. Expand the menu for the remote folder and select Use Dev Container — the workspace label changes to include the - Dev Container suffix.
4. Choose an available agent harness, configure the session, and enter your prompt.
VS Code adds subagent delegation with configurable harnesses, custom agent files, nested recursion, and per-agent model selection.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ USE IT
Create a read-only codebase research subagent that the main agent can delegate to without cluttering the main conversation.
.github/agents/codebase-researcher.agent.md
copy
# .github/agents/codebase-researcher.agent.md
---
description: Find relevant code and explain existing patterns
user-invocable: false
subagent: true
---
Research the requested topic without changing files.
Return relevant file paths, existing patterns, and unanswered questions.
Enable AI credit usage visibility and Auto model routing for subagents in VS Code settings.
📍 In VS Code, open Settings (Ctrl+,), search for subagents , then enable chat.subagents.showCreditUsage to surface credit costs per subagent response, and enable chat.subagents.defaultToAuto to route subagents through Auto model selection when no explicit model is configured.
Adds Run Subagent tool toggle in Configure Tools to enable subagent delegation in Local sessions.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Adds user-invocable and subagent frontmatter properties to custom agent .agent.md files, replacing the deprecated property that previously controlled both behaviors. Not checked yet Adds agents frontmatter property to coordinator agent files to restrict which workers they can invoke — agents: ['Codebase Researcher', 'Reviewer'] permits only named agents, agents: ['*'] permits all, agents: [] prevents subagent use. Not checked yet Adds model frontmatter property to custom agent files, accepting a model name or prioritized list, to pin subagents to a specific model independently of the main conversation. Not checked yet Adds chat.subagents.defaultToAuto setting to use Auto model selection for subagents when no model is specified by the tool call or agent configuration; defaults to false. Not checked yet
+5 more
show less
Adds nested subagent support for Local sessions, configurable via the nested subagent setting (disabled by default), with a maximum recursion depth of five. Not checked yet Adds chat.subagents.showCreditUsage setting to display AI credit usage in subagent response pills, hover details, and screen reader labels; hidden by default. Not checked yet Supports subagent delegation across four harnesses — Local, Copilot, Claude, and Codex — each with harness-specific configuration and behavior. Not checked yet In the Chat view (Local sessions), running subagents appear as collapsible tool calls showing agent name and current activity, with the prompt, tool calls, and result inspectable on selection. Not checked yet In the Agents window (supported sessions), subagents appear as read-only peer chats with a parent-chat indicator showing model, elapsed time, and active tool call; accessible from the dropdown, running-subagents indicator, or the in-transcript subagent pill. Not checked yet
10 details in Everything
Agents window gains Dev Container Dev Container A Docker-based development environment defined in a JSON config file, standardized by Microsoft, that lets a project ship its exact toolchain so any contributor's editor or CI system gets an identical workspace. sessions, Agent Merge, PR artifact removal, and a Create PR form with auto-merge controls.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ HOW TO FIND IT
Run an agent session inside your project's Dev Container so the agent has access to the exact build tools and dependencies defined in your repo.
📍 1. Open VS Code Settings and enable chat.agentHost.devContainer.enabled.
2. Open the Agents window and select your workspace at the top of the sidebar.
3. Expand the folder menu in the workspace picker and select Use Dev Container — the workspace label gains the - Dev Container suffix.
4. Choose your agent harness, configure the session, enter your prompt, and start the session.
Create a pull request from an Agent Host session with auto-merge and Agent Merge configured to fix CI failures automatically.
📍 1. In the Agents window, open the Pull Requests view and select Create PR .
2. Review repository, source branch, base branch, title, and description.
3. Under After creation , select Auto-Merge and choose a merge method (e.g. Squash ).
4. Under Blockers , select Fix CI Failures ; under Merge Pull Request , select When Ready .
5. Select Create PR — uncommitted changes are committed and the branch is pushed before the PR is created.
Attach a GitHub issue as context to a session by pasting its URL, so the agent has full issue details without manual copy-paste.
📍 Paste the GitHub issue URL (e.g. https://github.com/owner/repo/issues/42) directly into the Agents window prompt — VS Code automatically adds the issue as a context attachment.
Enables chat.agentHost.devContainer.enabled to run Agent Host sessions inside a Dev Container, supporting local folders and — starting in VS Code 1.139 — folders on SSH, Tunnel, or WSL WSL Windows Subsystem for Linux, a Microsoft-built compatibility layer that lets a Linux environment run directly on Windows without a virtual machine, giving cyber tools access to Linux binaries, filesystems, and syscalls on a Windows host. hosts.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Adds Use Dev Container option in the workspace picker so the agent builds and tests using the project's container tools and dependencies, with a - Dev Container suffix on the workspace label. Not checked yet Introduces Agent Merge, an experimental feature that monitors the pull request in an active agent session and asks the agent to address blockers, with configurable Blockers (Address Reviews , Fix CI Failures , Resolve Conflicts and Behind Branches ) and Merge Pull Request settings (Off , If Unchanged , When Ready ). Not checked yet Adds Agent Merge Changes changeset in the diff view, comparing the latest completed Agent Merge repair turn with the preceding completed user turn. Not checked yet Adds Mark Ready action to promote a draft pull request to ready for review while Agent Merge is active. Not checked yet
+1 more
show less
Supports attaching GitHub issues and pull requests as context via Add Context > Issue... or Pull Request..., or by pasting a GitHub issue or pull request URL directly into the prompt. Not checked yet
6 details in Everything
code-oss: Enables chat.agentHost.devContainer.enabled to run Agent Host sessions inside a Dev Container Dev Container A Docker-based development environment defined in a JSON config file, standardized by Microsoft, that lets a project ship its exact toolchain so any contributor's editor or CI system gets an identical workspace. so the agent builds and tests with the project's own tools and dependencies.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ HOW TO FIND IT
Manually enable Dev Container sessions when the gradual rollout has not yet reached your instance.
📍 1. Open VS Code Settings (Ctrl+, / Cmd+,).
2. Search for chat.agentHost.devContainer.enabled.
3. Set the toggle to enabled.
4. Open the Agents window, expand the workspace-picker menu for a local folder or a configured SSH/Tunnel/WSL folder, and select Use Dev Container to start the session inside the container.
Set up an Agent Merge workflow so the agent continuously addresses CI failures and review comments until the PR is ready to merge automatically.
📍 1. In the Agents window, open a session associated with a pull request (or use the Create PR form to create one).
2. In the Create PR form, under After creation , select Agent Merge .
3. Under Blockers , check Address Reviews and Fix CI Failures .
4. Under Merge Pull Request , select When Ready .
5. Select Create PR — Agent Merge begins monitoring and dispatches repair turns until required checks and approvals pass.
Attach a specific GitHub issue as context to an agent session without leaving the prompt.
📍 1. In the Agents window chat input, paste a GitHub issue URL (e.g. https://github.com/org/repo/issues/42) directly into the prompt.
2. VS Code automatically attaches the issue as a context item — no need to use the 'Add Context › Issue...' picker.
Adds a Create PR form for Agent Host sessions, with fields for repository, source branch, base branch, title, and description, and a Create as Draft toggle. Not checked yet Adds After creation options to the Create PR form — Merge Manually , Auto-Merge (with Squash , Merge Commit , or Rebase methods), and Agent Merge — as mutually exclusive post-creation behaviours. Not checked yet
2 details in Everything
VS Code Agents window gains per-diff-editor word wrap control and GitHub PR review comment support.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ HOW TO FIND IT
Stop long diff lines from scrolling horizontally in the Agents window without affecting your main code editor wrap setting.
📍 In the Agents window diff editor, select Word Wrap. This updates sessions.diffEditor.wordWrap independently of the editor.wordWrap setting.
Leave a PR review comment on a specific code range from the Agents window without asking the agent to revise the code.
📍 1. Enable chat.experimental.agentsWindowPRComments in VS Code settings.
2. Open a changed file in the Agents window view and select a range of lines that are part of the pull request diff.
3. Select Add PR Comment and enter your comment to post it directly to GitHub (shown as owner/repository#number when multiple PRs are associated).
Enables adding GitHub pull request review comments directly from the Agents window view via the chat.experimental.agentsWindowPRComments setting and the Add PR Comment action. Not checked yet
1 detail in Everything
VS Code Agents adds automatic session cleanup settings and an application-icon badge for unread session alerts.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ USE IT
Opt in to the application-icon badge on Stable so your team always sees how many Agent sessions need attention without opening VS Code.
yaml
copy
sessions.showApplicationBadge: true
Adds chat.agentSessions.autoMarkAsDoneMergedSessionsAfterDays config setting to automatically mark inactive Agent Host sessions as done after a configured number of days when their pull requests have merged.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Adds chat.agentSessions.autoDeleteArchivedMergedSessionsAfterDays config setting to permanently delete sessions that were automatically marked as done after a separate grace period (in days). Not checked yet Adds sessions.showApplicationBadge setting (defaults to true in Insiders, disabled in Stable) to show a badge on the OS application icon — macOS dock, Linux launcher, or Windows taskbar — counting Agent sessions with unread results, pending input, or failing CI checks on open non-draft pull requests. Not checked yet Surfaces a Configure Automatic Cleanup action in the Agents window when a merged pull request makes a session eligible, opening both cleanup settings without enabling them. Not checked yet
4 details in Everything
VS Code Agents gains sessions.useWorktree, chat.editor.preferCopilotHarness, and ChatEditorPreferCopilotHarness policy for harness and isolation control.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ HOW TO FIND IT
Enforce the Copilot SDK harness for all editor-chat sessions in VS Code so agents consistently use shared SDK Policy Hooks rather than Local hooks.
📍 Set chat.editor.preferCopilotHarness to true in VS Code settings (File › Preferences › Settings, search for chat.editor.preferCopilotHarness, and enable it).
Enables ~/.bashrc (macOS/Linux) or PowerShell profiles (Windows) before each command in local Copilot Agent Host sessions that use the SDK built-in shell tool. Not checked yet
1 detail in Everything
code-oss: Enables chat.agentHost.devContainer.enabled to run Agent Host sessions inside a Dev Container Dev Container A Docker-based development environment defined in a JSON config file, standardized by Microsoft, that lets a project ship its exact toolchain so any contributor's editor or CI system gets an identical workspace. , supporting local folders and — from VS Code 1.139 — folders on SSH, Tunnel, or WSL WSL Windows Subsystem for Linux, a Microsoft-built compatibility layer that lets a Linux environment run directly on Windows without a virtual machine, giving cyber tools access to Linux binaries, filesystems, and syscalls on a Windows host. remote hosts.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ HOW TO FIND IT
Enable Dev Container sessions so the agent can build and test inside your project's containerized environment before starting a session.
📍 In VS Code Settings, search for chat.agentHost.devContainer.enabled and set it to true. Then in the Agents window sidebar, expand the folder menu for your project and select Use Dev Container . The workspace label updates to show - Dev Container. Choose your agent harness and start the session.
Attach a GitHub issue as context so the agent has full issue details when working on a fix, without leaving the prompt.
📍 In the Agents window prompt, paste the GitHub issue URL directly (e.g. https://github.com/org/repo/issues/42). VS Code automatically adds it as a context attachment. Alternatively, select Add Context › Issue... from the toolbar and pick a repository and issue.
Create a pull request with Agent Merge enabled so the agent automatically addresses CI failures and review feedback until the PR is ready to merge.
📍 In the Agents window, open the Create PR form. Fill in the title and description, select Auto-Merge with merge method Squash , then under Agent Merge set Blockers to Fix CI Failures and Merge Pull Request to When Ready . Select Create PR . Agent Merge will monitor the PR and ask the agent to fix failures each cycle; track progress via Agent Merge Changes in the diff view.
Adds Use Dev Container and Use Local / Use Remote Host workspace-picker options so you can toggle Dev Container mode per folder before starting a session; the workspace label gains a - Dev Container suffix when active. Not checked yet Requires a .devcontainer/devcontainer.json Dev Container configuration at the project root and Docker installed and on PATH on the machine (or remote host) that contains the folder. Not checked yet Adds Remove Pull Request Artifact from Session via right-click on the pull request pill (single PR) or via the pull requests dropdown (multiple PRs), disassociating the artifact without closing the PR on GitHub. Not checked yet Adds a Create PR form for Agent Host sessions that lets you set repository, source branch, base branch, title, description, draft status, and post-creation behavior — including Merge Manually , Auto-Merge (with Squash , Merge Commit , or Rebase method), and Agent Merge options (Address Reviews , Fix CI Failures , Resolve Conflicts and Behind Branches ) plus Merge Pull Request modes Off , If Unchanged , and When Ready . Not checked yet Adds Send Create PR Message under Pull Request Actions in the Create PR form to send title, description, draft status, and merge choice to the session chat instead of creating the PR directly. Not checked yet
+2 more
show less
Adds a Windows jump-list shortcut: right-click the VS Code taskbar icon and select Agents Window from the Tasks jump list. Not checked yet Opens .html files in the integrated browser by default during an active desktop session, while HTML diffs continue to open in the diff editor. Not checked yet
Adds Use Dev Container and Use Local / Use Remote Host workspace-picker options so you can toggle Dev Container mode per folder before starting a session; the workspace label gains a - Dev Container suffix when active. Not checked yet Requires a .devcontainer/devcontainer.json Dev Container configuration at the project root and Docker installed and on PATH on the machine (or remote host) that contains the folder. Not checked yet Adds Remove Pull Request Artifact from Session via right-click on the pull request pill (single PR) or via the pull requests dropdown (multiple PRs), disassociating the artifact without closing the PR on GitHub. Not checked yet Adds a Create PR form for Agent Host sessions that lets you set repository, source branch, base branch, title, description, draft status, and post-creation behavior — including Merge Manually , Auto-Merge (with Squash , Merge Commit , or Rebase method), and Agent Merge options (Address Reviews , Fix CI Failures , Resolve Conflicts and Behind Branches ) plus Merge Pull Request modes Off , If Unchanged , and When Ready . Not checked yet Adds Send Create PR Message under Pull Request Actions in the Create PR form to send title, description, draft status, and merge choice to the session chat instead of creating the PR directly. Not checked yet
+1 more
show less
Adds a Windows jump-list shortcut: right-click the VS Code taskbar icon and select Agents Window from the Tasks jump list. Not checked yet
1 more in Everything
code-oss: MCP MCP Model Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. servers defined in .vscode/mcp.json and workspace-root .mcp.json inherit Workspace Trust, allowing trust decisions to be shared across related boundaries.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
code-oss: Adds chat.agent.sandbox.allowNetwork to toggle outbound network access for sandboxed terminal commands; defaults to true (unrestricted).
└──▷ USE IT
Lock down agent terminal commands on Linux to block all outbound network except the GitHub API, while granting read access to an app config dir and blocking SSH key access.
json
copy
{
"chat.agent.sandbox.enabled": true,
"chat.agent.sandbox.fileSystem.linux": {
"allowRead": ["/home/me/.config/myapp"],
"denyRead": ["/home/me/.ssh"]
},
"chat.agent.sandbox.allowNetwork": false,
"chat.agent.allowedNetworkDomains": ["api.github.com"],
"chat.agent.deniedNetworkDomains": []
}
Check exactly which sandbox restrictions are in effect for the current Agent Host session, or investigate why a terminal command was blocked.
Require manual approval for all sandboxed terminal commands and prevent the agent from falling back to unsandboxed execution.
json
copy
{
"chat.agent.sandbox.enabled": true,
"chat.agent.sandbox.allowAutoApprove": false,
"chat.agent.sandbox.allowUnsandboxedCommands": false
}
Adds chat.agent.allowedNetworkDomains and chat.agent.deniedNetworkDomains for domain-level network filtering in Local sessions and Agent Host custom terminal tool on macOS, Linux, and WSL2 WSL2 A Microsoft-maintained compatibility layer in Windows 10/11 that runs a real Linux kernel inside a lightweight VM, letting cyber tools built for Linux execute natively on Windows without dual-booting or a full VM. . Not checked yet Adds chat.agentHost.sdkSandbox.enabled (macOS, Linux, WSL2) and chat.agentHost.sdkSandbox.enabledWindows (Windows) to enable Experimental sandboxing sandboxing Running code inside a restricted environment that limits what it can access or change. The isolation reduces the impact of mistakes or untrusted code, depending on the sandbox's controls. for the Copilot SDK built-in shell used in Agent Host sessions. Not checked yet Adds chat.agentHost.customTerminalTool.enabled to switch Agent Host sessions from the Copilot SDK built-in shell to the custom terminal tool (which uses the chat.agent.sandbox.* settings); defaults to false. Not checked yet Adds /sandbox-policy slash command in Copilot Agent Host sessions to inspect the effective sandbox policy — showing whether sandboxing is active, the OS sandbox implementation, and the file system and network rules in effect. Not checked yet
4 details in Everything
VS Code 1.139 adds remote Dev Container Dev Container A Docker-based development environment defined in a JSON config file, standardized by Microsoft, that lets a project ship its exact toolchain so any contributor's editor or CI system gets an identical workspace. support for agent sessions over SSH, Tunnel, and WSL WSL Windows Subsystem for Linux, a Microsoft-built compatibility layer that lets a Linux environment run directly on Windows without a virtual machine, giving cyber tools access to Linux binaries, filesystems, and syscalls on a Windows host. hosts.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
VS Code Copilot gains a dedicated /plan workflow with Plan mode, review cards, and Local session Plan agent controls.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ TRY IT
Kick off a scoped planning request in Copilot chat so the agent researches your project and drafts a reviewable plan before touching any code.
copy
$ /plan Add pagination to this project's product-list API. Follow the existing API conventions and preserve the behavior of requests that don't specify pagination. Include tests for invalid parameters, empty results, and compatibility with existing clients. Identify decisions I need to make before implementation.
Allow the Agent to self-select Plan mode when a task is complex, reducing manual mode-switching for research-heavy requests.
json
copy
github.copilot.chat.switchAgent.enabled: true
Review and approve a finished plan without immediately executing it, then switch models before asking the agent to implement.
📍 1. In the Review Plan card, select Open Full Plan and review the Markdown plan file, adding inline comments as needed.
2. Select Submit Feedback to send revisions to the agent and wait for the updated plan.
3. Select Approve Plan Only to approve without starting implementation.
4. Switch the model in the language model picker to a lower-cost model.
5. Switch from Plan to Interactive in the agent mode picker, then ask the agent to implement the approved plan.
Adds /plan slash command in Copilot chat to switch the session into Plan mode and submit a planning request in one step.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Adds github.copilot.chat.switchAgent.enabled setting to let the Agent automatically switch to Plan mode for research or complex tasks. Not checked yet Introduces a dedicated Plan mode selectable from the agent mode picker, enabling the agent to research the project, ask clarifying questions, and produce a reviewable plan without editing files. Not checked yet Adds a Review Plan card with Open Full Plan , Submit Feedback , Implement Plan , Approve Plan Only , Implement with Autopilot , and Reject actions to control plan approval and execution separately from tool-action approval. Not checked yet Adds Open in Editor action in Local sessions that switches to Agent mode and creates an editable, saveable copy of the plan. Not checked yet
+4 more
show less
Introduces session handoff via a Continue in destination dropdown next to the implement action, allowing a plan to be handed off to another agent session. Not checked yet Adds Start with Autopilot Preview action in Local sessions to begin implementation with automatic tool approval, covering file edits, terminal commands, and external tool calls. Not checked yet Supports a built-in Plan agent in the Local session target, accessible by selecting Local from the Session Target control, with read-only project research by default. Not checked yet Exposes Local planning settings including a model picker for planning, an experimental tool-addition setting, and the github.copilot.chat.switchAgent.enabled toggle. Not checked yet
9 details in Everything
VS Code chat gains terminal command execution, request queuing/steering, OS notifications, find-in-chat, and a customizable Agents window background.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ HOW TO FIND IT
Stay aware of long-running agent tasks completing while you work elsewhere in VS Code, without switching back to check.
📍 In VS Code Settings, set chat.notifyWindowOnResponseReceived to always and chat.notifyWindowOnConfirmation to always.
Run your test suite from the Chat view without involving the agent — useful for quickly checking results mid-conversation.
Change the default queuing behaviour so new messages wait rather than steering the current request.
📍 In VS Code Settings, set chat.requestQueuing.defaultAction to queue.
Adds chat.requestQueuing.defaultAction setting to control whether queued messages default to steer or queue when a request is already in progress.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Adds chat.notifyWindowOnResponseReceived setting (values: off, windowNotFocused, always) to send OS notifications when a chat response is received, with a preview of the response. Not checked yet Adds chat.notifyWindowOnConfirmation setting (values: off, windowNotFocused, always) to send OS notifications when the agent needs user input or confirmation to continue. Not checked yet Adds imageCarousel.chat.enabled setting (set to false to disable) to control the new image/video carousel view for media in chat responses. Not checked yet Adds chat.fontFamily and chat.fontSize settings to control the markdown font in the chat transcript. Not checked yet
+10 more
show less
Adds chat.editor.fontFamily, chat.editor.fontSize, chat.editor.fontWeight, chat.editor.lineHeight, and chat.editor.wordWrap settings to control code block font and layout in chat. Not checked yet Adds chat.stickyScroll.enabled setting to pin the current prompt to the top of the transcript while scrolling. Not checked yet Adds chat.verbose setting to show or hide per-request timestamps and elapsed response time in chat. Not checked yet Enables running terminal commands directly from the Chat view or Agents window by starting a message with ! (Agent Host sessions only), executing immediately in the session's default shell without agent involvement. Not checked yet Adds three in-flight message handling options — Add to Queue , Steer with Message , and Stop and Send — via the Send button dropdown while a request is in progress. Not checked yet Supports drag-and-drop reordering of pending (queued or steering) messages in the chat input queue. Not checked yet Adds full-text search across an entire chat session (prompts, responses, code blocks, collapsed steps) via ⌘F / Ctrl+F , available in the Chat view, chat editor tabs, and Agents window. Not checked yet Adds keyboard shortcuts to navigate between prompts (⌥⌘↑/Ctrl+Alt+Up , ⌥⌘↓/Ctrl+Alt+Down ) and code blocks (⌥⌘PageUp/Ctrl+Alt+PageUp, ⌥⌘PageDown/Ctrl+Alt+PageDown) within a chat session. Not checked yet Adds the /vscode-pet slash command to show or hide an interactive pet above the chat input box, with drag, throw, hop, and achievement interactions. Not checked yet Adds Chat: Set Background... and Chat: Change Background Layout... Command Palette commands to set a decorative background (none, Codicons pattern, or custom image) in the Agents window chat area, with per-theme and per-layout persistence. Not checked yet
15 details in Everything
└──▷ USE IT
Force all code editors in the Agents window to wrap at the viewport width, regardless of the global editor word-wrap setting.
json
copy
sessions.editor.wordWrap: on
VS Code Insiders adds sessions.useWorktree setting and chat.editor.preferCopilotHarness to control agent harness agent harness The software around an AI agent that manages its tools, execution loop, state and limits. It determines how the model's proposed actions are run and checked. selection︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ USE IT
Enforce the Copilot SDK harness for all editor-chat sessions org-wide so agents consistently use shared SDK hooks and Copilot Policy Hooks.
yaml
copy
In VS Code settings (or your org device policy), set:
chat.editor.preferCopilotHarness: true
Enable isolated Git worktree sessions by default for all new workspace sessions in VS Code Insiders to prevent cross-session file interference.
yaml
copy
In VS Code Insiders settings, set:
sessions.useWorktree: true
Adds sessions.useWorktree setting in VS Code Insiders to control whether the isolated Git worktree harness is selected when creating a first workspace session.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Adds chat.editor.preferCopilotHarness setting to prefer the Copilot SDK harness over Local for new editor-chat sessions. Not checked yet Adds ChatEditorPreferCopilotHarness device policy (available from version 1.134) for enterprise admins to enforce the Copilot harness preference. Not checked yet Supports Dev Container Dev Container A Docker-based development environment defined in a JSON config file, standardized by Microsoft, that lets a project ship its exact toolchain so any contributor's editor or CI system gets an identical workspace. execution in the desktop Agents window, selectable via the workspace picker for local, SSH, Tunnel, or WSL WSL Windows Subsystem for Linux, a Microsoft-built compatibility layer that lets a Linux environment run directly on Windows without a virtual machine, giving cyber tools access to Linux binaries, filesystems, and syscalls on a Windows host. -hosted projects. Not checked yet Enables agent session handoff from a Local session to other supported sessions via a Continue in dropdown destination. Not checked yet
5 details in Everything
code-oss: Enables chat.agentHost.devContainer.enabled to run Agent Host sessions inside a Dev Container Dev Container A Docker-based development environment defined in a JSON config file, standardized by Microsoft, that lets a project ship its exact toolchain so any contributor's editor or CI system gets an identical workspace. , letting the agent build and test with the project's own tools and dependencies.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ USE IT
Opt in to Dev Container sessions so agent runs have access to your project's exact toolchain rather than the host environment.
yaml
copy
chat.agentHost.devContainer.enabled: true
Start a Dev Container agent session on a remote SSH folder — in the Agents window sidebar, expand the folder menu for the SSH-connected folder and choose Use Dev Container , then pick an agent harness agent harness The software around an AI agent that manages its tools, execution loop, state and limits. It determines how the model's proposed actions are run and checked. and enter your prompt.
📍 In the Agents window sidebar:
1. Expand the folder menu for your SSH/Tunnel/WSL folder.
2. Select Use Dev Container — the label gains a - Dev Container suffix.
3. Choose an agent harness, configure the session, and enter your prompt.
To revert before starting: expand the folder menu and select Use Remote Host .
Create a draft PR and immediately start Agent Merge so the agent keeps resolving CI and review blockers autonomously while you work on other things.
📍 In the session view:
1. Open the pull request action menu.
2. Select Create Draft PR & Agent Merge .
3. Configure which blockers Agent Merge should address.
4. When ready for review, select Mark Ready from the action menu.
Supports Dev Container sessions on remote folders over SSH, Tunnel, or WSL WSL Windows Subsystem for Linux, a Microsoft-built compatibility layer that lets a Linux environment run directly on Windows without a virtual machine, giving cyber tools access to Linux binaries, filesystems, and syscalls on a Windows host. hosts (requires VS Code 1.139), with Docker running on the remote host and a .devcontainer/devcontainer.json present. Not checked yet Adds GitHub Issues and Pull Requests as first-class context attachments via 'Add Context > Issue...' or 'Pull Request...', with direct URL paste support. Not checked yet
2 details in Everything
Custom instructions now support ~ home-folder paths and Unix-style / separators for cross-OS portability.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ USE IT
Reference a personal coding standards file from your home folder inside a custom instructions file, so the same instructions file works on any OS.
text
copy
[personal coding standards](~/copilot/coding-standards.md)
Supports ~ and ~/ prefix in custom instructions file paths to reference the user home folder, e.g.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ [personal coding standards](~/copilot/coding-standards.md). Not checked yet Supports #file: syntax and Markdown links in custom instructions to reference specific files or URLs as context, with relative paths resolving from the instructions file. Not checked yet
Supports ~ and ~/ prefix in custom instructions file paths to reference the user home folder, e.g.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ [personal coding standards](~/copilot/coding-standards.md). Not checked yet
1 more in Everything
code-oss adds sessions.diffEditor.wordWrap to control line wrapping in diff editors within the Agents window.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ USE IT
Force diff editors in the Agents window to always wrap long lines, making wide diffs readable without horizontal scrolling.
yaml
copy
sessions.diffEditor.wordWrap: on
VS Code agent terminal sandboxing sandboxing Running code inside a restricted environment that limits what it can access or change. The isolation reduces the impact of mistakes or untrusted code, depending on the sandbox's controls. adds OS-level file system and network isolation for agent commands on macOS, Linux, WSL2 WSL2 A Microsoft-maintained compatibility layer in Windows 10/11 that runs a real Linux kernel inside a lightweight VM, letting cyber tools built for Linux execute natively on Windows without dual-booting or a full VM. , and Windows Experimental .︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ USE IT
Lock down an agent session so terminal commands can read an app config dir but cannot touch your SSH keys or reach any network host except the GitHub API.
json
copy
{
"chat.agent.sandbox.enabled": true,
"chat.agent.sandbox.fileSystem.linux": {
"allowRead": ["/home/me/.config/myapp"],
"denyRead": ["/home/me/.ssh"]
},
"chat.agent.sandbox.allowNetwork": false,
"chat.agent.allowedNetworkDomains": ["api.github.com"],
"chat.agent.deniedNetworkDomains": []
}
Harden the sandbox so a blocked network call never silently retries unsandboxed and sandboxed commands still go through the normal approval prompt.
json
copy
{
"chat.agent.sandbox.enabled": true,
"chat.agent.sandbox.allowNetwork": false,
"chat.agent.sandbox.retryWithAllowNetworkRequests": false,
"chat.agent.sandbox.allowUnsandboxedCommands": false,
"chat.agent.sandbox.allowAutoApprove": false
}
Enable sandboxing on Windows Experimental after applying the required September 2026 security update.
📍 In VS Code Settings, search for chat.agent.sandbox.enabledWindows and set it to on, or select Sandboxing for terminal in the permissions picker. Confirm the shield indicator is active before running agent terminal commands.
Enables agent terminal sandboxing via chat.agent.sandbox.enabled (macOS, Linux, WSL2) or chat.agent.sandbox.enabledWindows (Windows, Experimental), defaulting to off.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Adds chat.agent.sandbox.fileSystem.mac, chat.agent.sandbox.fileSystem.linux, and chat.agent.sandbox.fileSystem.windows settings to configure per-platform file system access rules (allowRead, allowWrite, denyRead, denyWrite) using literal paths or Git-style globs (macOS only). Not checked yet Adds chat.agent.sandbox.allowNetwork (default true) to toggle network isolation independently of file system restrictions. Not checked yet Adds chat.agent.allowedNetworkDomains and chat.agent.deniedNetworkDomains to define the outbound network allowlist/denylist for sandboxed terminal commands; denied domains take precedence. Not checked yet Adds chat.agent.networkFilter to extend the same domain allowlist/denylist to the fetch tool and integrated browser (requires VS Code restart after changes). Not checked yet
+5 more
show less
Adds chat.agent.sandbox.retryWithAllowNetworkRequests (default true) to control whether the agent can prompt the user to retry a network-blocked sandboxed command with unrestricted network access. Not checked yet Adds chat.agent.sandbox.allowAutoApprove to control whether sandboxed terminal commands are auto-approved or follow the normal terminal approval flow. Not checked yet Adds chat.agent.sandbox.allowUnsandboxedCommands (default true) to control whether the agent can fall back to running a command outside the sandbox when it cannot execute inside it. Not checked yet Surfaces a shield indicator in the permissions picker showing the effective sandbox state for the current session. Not checked yet Supports per-session sandboxing toggle in Copilot Agent Host sessions; peer chats and subagents inherit the owning session's sandbox state. Not checked yet
10 details in Everything
VS Code agent sandboxing sandboxing Running code inside a restricted environment that limits what it can access or change. The isolation reduces the impact of mistakes or untrusted code, depending on the sandbox's controls. (Preview on macOS/Linux/WSL2 WSL2 A Microsoft-maintained compatibility layer in Windows 10/11 that runs a real Linux kernel inside a lightweight VM, letting cyber tools built for Linux execute natively on Windows without dual-booting or a full VM. , Experimental on Windows) restricts file system and network access for terminal commands.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ HOW TO FIND IT
After changing the network policy setting, restart VS Code to ensure new integrated browser sessions pick up the updated policy.
📍 In VS Code, open Settings and update your agent sandboxing User or Workspace setting, then restart VS Code to apply the new network policy to integrated browser sessions.
Provides a sandboxing toggle in the permissions picker to enable or disable sandboxing per session for Copilot Agent Host sessions, with new sessions inheriting the effective User or Workspace setting.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet
1 detail in Everything
VS Code Agents window gains Dev Container Dev Container A Docker-based development environment defined in a JSON config file, standardized by Microsoft, that lets a project ship its exact toolchain so any contributor's editor or CI system gets an identical workspace. sessions, GitHub issue/PR context, and experimental Agent Merge for automated PR unblocking.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ HOW TO FIND IT
Run an agent session inside a project's Dev Container to give the agent access to the exact toolchain and dependencies the project requires.
📍 1. Open VS Code Settings and enable chat.agentHost.devContainer.enabled.
2. Open the Agents window from the sidebar.
3. In the workspace picker, expand the menu for a local folder that has a .devcontainer/devcontainer.json and select Use Dev Container .
4. Choose an agent harness (Copilot harness or Codex in Interactive mode), configure the session, and enter your prompt.
Enable Agent Merge to automatically unblock a pull request — create a draft PR and start monitoring in one step.
📍 1. Open the session view for an agent session that does not yet have a pull request.
2. Open the pull request action menu and select Create Draft PR & Agent Merge .
3. Configure which blockers Agent Merge should address.
4. When checks and review feedback are clear, select Mark Ready from the action menu to promote the draft PR to ready-for-review.
Attach a GitHub issue or pull request as context so the agent can reference it during a session.
📍 1. In the Agents window prompt area, select 'Add Context › Issue...' or 'Add Context › Pull Request...' (if multiple repositories are available, select the repository first).
2. Alternatively, paste a GitHub issue or pull request URL directly into the prompt — VS Code automatically adds it as a context attachment.
Enables chat.agentHost.devContainer.enabled (off by default) to run an Agent Host session inside a Dev Container, giving the agent access to the project's tools, dependencies, and environment defined in .devcontainer/devcontainer.json.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Adds Use Dev Container workspace option in the Agents window sidebar; the workspace label gains a - Dev Container suffix, and Use Local reverts the selection before the session starts. Not checked yet Supports adding GitHub issues and pull requests as context via 'Add Context > Issue...' or 'Pull Request...' in the prompt, or by pasting a GitHub issue or pull request URL directly into the prompt. Not checked yet Introduces experimental Agent Merge, which monitors the pull request associated with an agent session and asks the agent to address blockers until the pull request is ready to merge. Not checked yet Adds Create Draft PR & Agent Merge action to create a draft pull request and start Agent Merge monitoring in a single step from the session view. Not checked yet
+2 more
show less
Adds Agent Merge Changes changeset in the session view, comparing the latest completed Agent Merge repair turn with the preceding completed user turn. Not checked yet Adds Mark Ready action in the session view to promote a draft pull request to ready-for-review while Agent Merge is active. Not checked yet
7 details in Everything
code-oss gains automatic Agent Host session cleanup settings and an application-icon badge for session attention counts.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ USE IT
Automatically retire and delete Agent Host sessions whose PRs have merged, preventing stale sessions from accumulating over time.
json
copy
// In VS Code settings (settings.json)
{
"chat.agentSessions.autoMarkAsDoneMergedSessionsAfterDays": 7,
"chat.agentSessions.autoDeleteArchivedMergedSessionsAfterDays": 7
}
Disable the application-icon session-attention badge in VS Code Insiders if you prefer a clean taskbar/dock.
json
copy
// In VS Code settings (settings.json)
{
"sessions.showApplicationBadge": false
}
Adds chat.agentSessions.autoMarkAsDoneMergedSessionsAfterDays setting to automatically mark inactive merged Agent Host sessions as done after a configurable number of days.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Adds chat.agentSessions.autoDeleteArchivedMergedSessionsAfterDays setting to permanently delete automatically-completed merged sessions after a separate configurable grace period. Not checked yet Adds sessions.showApplicationBadge setting (defaults to true in Insiders, disabled in Stable) to show a dock/launcher/taskbar badge counting Agent sessions that have unread results, are waiting for input, or have failing CI checks on an open non-draft PR. Not checked yet
3 details in Everything
VS Code adds agent sandboxing sandboxing Running code inside a restricted environment that limits what it can access or change. The isolation reduces the impact of mistakes or untrusted code, depending on the sandbox's controls. (Preview/Experimental) to restrict file system and network access for agent-executed terminal commands.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
VS Code adds OS-level agent terminal sandboxing sandboxing Running code inside a restricted environment that limits what it can access or change. The isolation reduces the impact of mistakes or untrusted code, depending on the sandbox's controls. with configurable file system and network boundaries on macOS, Linux, WSL2 WSL2 A Microsoft-maintained compatibility layer in Windows 10/11 that runs a real Linux kernel inside a lightweight VM, letting cyber tools built for Linux execute natively on Windows without dual-booting or a full VM. , and Windows.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ USE IT
Lock down a sandboxed agent to only reach GitHub's API while blocking all other outbound network — for example, during an automated PR-review agent run where you want to prevent exfiltration to arbitrary hosts.
json
copy
{
"chat.agent.sandbox.enabled": true,
"chat.agent.sandbox.allowNetwork": false,
"chat.agent.allowedNetworkDomains": ["api.github.com"],
"chat.agent.deniedNetworkDomains": ["example.com"]
}
Grant a sandboxed Linux agent read access to an app config directory while explicitly blocking the SSH folder, preventing credential exposure during build or test runs.
json
copy
{
"chat.agent.sandbox.enabled": true,
"chat.agent.sandbox.fileSystem.linux": {
"allowRead": ["/home/me/.config/myapp"],
"denyRead": ["/home/me/.ssh"]
}
}
Require explicit approval for every sandboxed terminal command and disable the unsandboxed fallback, ensuring no agent command runs outside the OS-level boundary without user confirmation.
json
copy
{
"chat.agent.sandbox.enabled": true,
"chat.agent.sandbox.allowAutoApprove": false,
"chat.agent.sandbox.allowUnsandboxedCommands": false
}
Enables agent terminal sandboxing via chat.agent.sandbox.enabled (macOS, Linux, WSL2) or chat.agent.sandbox.enabledWindows (Windows), defaulting to off, with a shield indicator in the permissions picker showing effective state.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Adds chat.agent.sandbox.fileSystem.mac, chat.agent.sandbox.fileSystem.linux, and chat.agent.sandbox.fileSystem.windows settings to control sandbox file system access using allowRead, allowWrite, denyRead, and denyWrite rules. Not checked yet Adds chat.agent.sandbox.allowNetwork (default true) to toggle network isolation independently of file system restrictions for sandboxed terminal commands. Not checked yet Adds chat.agent.allowedNetworkDomains and chat.agent.deniedNetworkDomains to define per-domain network allow/deny lists for sandboxed commands, with denied domains taking precedence. Not checked yet Adds chat.agent.networkFilter to apply the same chat.agent.allowedNetworkDomains / chat.agent.deniedNetworkDomains domain lists to the fetch tool and integrated browser. Not checked yet
+5 more
show less
Adds chat.agent.sandbox.retryWithAllowNetworkRequests (default true) to control whether the agent can request a retry with unrestricted network access when a sandboxed command is blocked by network isolation. Not checked yet Adds chat.agent.sandbox.allowAutoApprove to control whether sandboxed terminal commands are auto-approved (default) or go through the normal terminal approval flow. Not checked yet Adds chat.agent.sandbox.allowUnsandboxedCommands (default true) to control whether the agent may fall back to running a command outside the sandbox when it cannot execute inside it. Not checked yet Supports per-session sandboxing toggle for Copilot Agent Host sessions, with explicit session selections persisting across reloads and restarts independently of User or Workspace settings. Not checked yet Sandboxing on Linux and WSL2 requires bubblewrap and socat; Windows Experimental requires the September 8, 2026 security update (KB5124008 for Windows 11 24H2/25H2, KB5124012 for Windows 11 26H1). Not checked yet
10 details in Everything
VS Code agent automations add .automation.md export/import, scheduling, and plugin-contributed templates︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ HOW TO FIND IT
Share a configured automation with a teammate by exporting it to a portable .automation.md file they can import on their own machine.
📍 On the automation card, select More Actions › Export, then choose where to save the proposed .automation.md file. The recipient selects Import Automation in the Automations view and chooses the file, then picks their own workspace, agent, model, and permissions.
Quickly set up a recurring task (such as a daily repo-change summary) by starting from a built-in or plugin-contributed template instead of writing a prompt from scratch.
📍 In the Automations view, expand Built-in Templates or Templates from Plugins , select a template, review the prompt and schedule in the New Automation dialog, choose workspace/agent/model/permissions, then enable and save.
Adds agent automations that run saved prompt-and-schedule tasks on demand or on a recurring (manual, hourly, daily, or weekly) schedule without starting each session manually.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Supports exporting an automation as a .automation.md file (readable Markdown with YAML frontmatter) to share or reuse across machines. Not checked yet Supports importing a .automation.md file via the Import Automation button in the Automations view, or by dragging the file onto the view. Not checked yet Adds automation templates sourced from built-in VS Code templates and from enabled Agent Plugins, with plugin-contributed templates marked with a Plugin badge and a blue marker on the Templates from Plugins section when new templates are available. Not checked yet Adds More Actions menu on automation cards with Enable /'Disable', Duplicate , Export , and Delete operations for managing saved automations. Not checked yet
5 details in Everything
Agent Host sessions can now run inside a Dev Container Dev Container A Docker-based development environment defined in a JSON config file, standardized by Microsoft, that lets a project ship its exact toolchain so any contributor's editor or CI system gets an identical workspace. , giving agents access to project-defined tools and dependencies.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ HOW TO FIND IT
Run an Agent Host session inside a Dev Container to give the agent access to project-specific tooling and dependencies defined in the container config.
📍 1. Enable the feature: set chat.agentHost.devContainer.enabled to true in VS Code settings.
2. Open the Agents sidebar and, in the workspace picker, expand the menu for a local folder that has a Dev Container configuration.
3. Select Use Dev Container — the workspace label changes to include the - Dev Container suffix.
4. Choose the Copilot harness (or Codex in Interactive mode), configure the session, and enter your prompt.
Adds a Use Dev Container option in the workspace picker for eligible local folders, with a - Dev Container suffix on the workspace label to indicate container mode.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Dev Container sessions support the Copilot harness and Codex (in Interactive mode) on the Agent Host, scoped to local folders only. Not checked yet
2 details in Everything
code-oss: File system boundary limits read and write access to configured paths, protecting sensitive locations such as SSH keys and shell configuration, with policies inherited by child processes like package managers and build scripts.
Network boundary limits outbound connections to configured domains to reduce the risk of data exfiltration data exfiltration Moving data out of a system or organization without authorization. It can happen through network transfers, removable media or other channels. and unintended actions on external services. Not checked yet On Windows, Microsoft MXC process containers apply file system and network policies to the command process (requires installing the applicable Windows security update). Not checked yet By default, terminal commands running inside the sandbox are automatically approved without a confirmation prompt, since they already execute in a controlled environment. Not checked yet
3 details in Everything
VS Code agent now supports user-level instruction files, recursive folder scanning, and a chat customization diagnostics view.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ HOW TO FIND IT
Use the diagnostics view to confirm your instruction files loaded correctly and spot any errors before a chat session.
📍 In the Chat view, right-click to open the context menu and select checking customization diagnostics to see all loaded instruction files and any errors.
Adds user-level instructions scope so instruction files apply across all workspaces, not just a single workspace.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Supports recursive scanning of instruction file folders, enabling subdirectory organization by team, language, or module. Not checked yet Adds a chat customization diagnostics view — accessible via the Chat view context menu — to inspect all loaded instruction files and surface any errors. Not checked yet Deprecates the previous instructions setting, which is now only used by the Local agent; existing customizations must be migrated to supported locations. Not checked yet
Adds user-level instructions scope so instruction files apply across all workspaces, not just a single workspace.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Adds a chat customization diagnostics view — accessible via the Chat view context menu — to inspect all loaded instruction files and surface any errors. Not checked yet
2 more in Everything
└──▷ BREAKING ON UPGRADE
! The previous instructions setting is deprecated and now only used by the Local agent; instruction file locations configured with it must be migrated to supported locations.
VS Code ships Agent Host for persistent, portable agent sessions plus an open Agent Host Protocol (AHP AHP Analytic Hierarchy Process, a structured decision-making method that breaks a complex choice into a weighted hierarchy of criteria and scores alternatives against them, letting cyber tools rank or prioritize options mathematically. ) with multi-language client libraries.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ TRY IT
Start a standalone Agent Host so external AHP clients (or a browser at insiders.vscode.dev/agents) can connect to your local sessions without keeping an editor window open.
Monitor a running agent session from the browser while the host continues work on your dev machine — navigate to the Agents window in VS Code, connect it to the remote Agent Host, start a session, then open the same host in the browser.
📍 Open the Agents window in VS Code › select your remote Agent Host from the host picker › start or open a session › navigate to insiders.vscode.dev/agents in your browser › select the same host to continue the session.
Adds code agent host (and code-insiders agent host for Insiders) CLI subcommand to start a standalone Agent Host process on your machine.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Introduces the Agent Host, a self-contained process that owns agent sessions independently of any editor window, keeping sessions alive after closing a folder or window. Not checked yet Introduces the open Agent Host Protocol (AHP) for coordinating multiple independent clients around the same long-running agent session via URI-addressable channels (sessions, chats, terminals, changesets). Not checked yet Publishes AHP client libraries in Rust (ahp, ahp-types, ahp-ws), TypeScript (@microsoft/agent-host-protocol), Kotlin (com.microsoft.agenthostprotocol:agent-host-protocol), Go (github.com/microsoft/agent-host-protocol/clients/go), and Swift (Swift Package Manager: microsoft/agent-host-protocol). Not checked yet Enables remote agent sessions where the Agent Host runs on a remote machine and clients connect from desktop, browser (insiders.vscode.dev/agents), SSH, or dev tunnels. Not checked yet
+2 more
show less
Supports multiple agent harnesses agent harnesses The software around an AI agent that manages its tools, execution loop, state and limits. It determines how the model's proposed actions are run and checked. (Copilot via GitHub Copilot SDK, Claude via Anthropic Claude Agent SDK) within a single Agent Host, each retaining its own agent loop while sharing a common AHP session model. Not checked yet Adds the Agents window as a synchronized client surface, letting practitioners monitor and interact with the same live session from both an editor window and the Agents window simultaneously. Not checked yet
7 details in Everything
VS Code Agent Customizations editor gains experimental migration flows and config keys to convert prompt files and location settings to Agent Host.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ USE IT
Enable location-settings migration so the Agent Customizations editor surfaces the Migrate Location Settings card for your Agent Host.
yaml
copy
chat.customizations.locationsMigration.enabled: true
Convert existing workspace prompt files to agent skills so they continue working after the Local agent is removed.
📍 1. In the Chat view, select the Agent Host that should use the skills.
2. Open the Agent Customizations editor and select Migrate Prompt Files .
3. Choose 'Convert to Skills...' and select the prompt files to convert.
4. In the confirmation dialog, choose whether to delete the original prompt files, then select Convert to Skills .
Adds chat.customizations.promptMigration.enabled setting (default true) to control prompt-file-to-skills migration in Agent Host sessions.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Adds chat.customizations.locationsMigration.enabled setting (default false) to migrate custom agents, instructions, and skills from locations configured via deprecated chat.agentFilesLocations, chat.modeFilesLocations, chat.instructionsFilesLocations, and chat.agentSkillsLocations settings. Not checked yet Adds Migrate Prompt Files and Migrate Location Settings migration cards in the Agent Customizations editor, surfaced when an Agent Host is selected and the corresponding setting is enabled. Not checked yet Agent Host sessions now load user-level customizations from supported folders instead of VS Code profile user data and VS Code-specific formats. Not checked yet
4 details in Everything
└──▷ BREAKING ON UPGRADE
! Prompt files are deprecated for Agent Host sessions and are not loaded by Agent Host; they must be converted to agent skills to remain available (the Local agent will be removed in a future release).! The chat.agentFilesLocations, chat.modeFilesLocations, chat.instructionsFilesLocations, and chat.agentSkillsLocations settings are deprecated because Agent Host sessions do not use them.! Custom agents and instructions stored in VS Code profile user data are not read by Agent Host; they must be migrated to the user folders for the selected Agent Host.
VS Code Agents window gains single-pane layout, Markdown editor, per-window settings overrides, and extension opt-in support.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ USE IT
Opt a specific extension into the Agents window so it activates during agent sessions alongside your main editor.
json
copy
{
"extensions.supportAgentsWindow": {
"myextension.id": true
}
}
Enable the experimental single-pane layout in the Agents window so files and diffs open in a docked editor next to chat instead of a modal.
json
copy
{
"sessions.layout.singlePaneDetailPanel": true
}
Allow Claude or BYOK BYOK Bring Your Own Key — a cloud deployment model where the customer supplies and controls the encryption keys rather than the provider, so sensitive data stays under the customer's cryptographic authority. models to run in the desktop Agents window without requiring GitHub sign-in.
json
copy
{
"chat.agentHost.allowSignedOutWhenUsable": true,
"chat.agentHost.byokModels.enabled": true
}
Adds chat.agentHost.byokModels.enabled setting to make Bring Your Own Key models available to Agent Host sessions in the Agents window.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Adds chat.agentHost.allowSignedOutWhenUsable setting to enable Claude, Codex, and BYOK models in the desktop Agents window without GitHub sign-in Experimental . Not checked yet Adds sessions.layout.singlePaneDetailPanel setting to enable an experimental single-pane layout that replaces the separate editor and side panel with one docked pane sharing a tab bar for Changes, Files, Browser, and Search. Not checked yet Adds sessions.layout.autoCollapseSessionsSidebar setting to automatically hide the sessions sidebar on narrow windows when both the editor area and side panel are open. Not checked yet Adds workbench.editor.markdownDefaultEditorInAgentsWindow setting to control whether .md files open with Markdown Editor Experimental (editing and locked modes with Git change markers) or Markdown Preview by default. Not checked yet
+3 more
show less
Adds extensions.supportAgentsWindow setting to opt VS Code extensions into the Agents window by extension ID. Not checked yet Supports per-window settings overrides so individual settings can be scoped to the Agents window without affecting the main VS Code editor window. Not checked yet Adds a Markdown Editor Experimental for .md files in the Agents window with Editing and Locked modes and Git change markers (green for added, blue for modified, red for deleted) in the margin. Not checked yet
8 details in Everything
VS Code agents gain layered approval controls, per-command terminal rules, and sandbox isolation for file system and network access︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ HOW TO FIND IT
Lock down a high-risk session by keeping the sandbox on but requiring explicit approval for every terminal command inside it — useful when reviewing untrusted code.
📍 In the chat input, open the permissions dropdown and select Manual permissions , then confirm that chat.agent.sandbox.enabled is true and chat.agent.sandbox.allowAutoApprove is false in your workspace settings.
Adds chat.agent.sandbox.allowAutoApprove setting to require normal terminal approval flow for commands running inside the sandbox.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Adds object syntax with matchFullCommand to evaluate the full command line instead of each subcommand individually in terminal auto-approval rules. Not checked yet Adds chat.agent.terminalApproval.useOnlyCustomRules experimental setting to ignore built-in terminal approval rules and rely solely on user-defined rules. Not checked yet Adds per-command terminal auto-approval rules in settings — set a command to true to auto-approve or false to require approval, and wrap patterns in / characters for regex matching. Not checked yet Adds chat.agent.sandbox.allowNetworkAccess setting to give sandboxed terminal commands unrestricted network access while keeping file system restrictions active. Not checked yet
+9 more
show less
Adds two-step URL approval separating request approval (via the Trusted Domains list) from response approval, preventing both data exfiltration data exfiltration Moving data out of a system or organization without authorization. It can happen through network transfers, removable media or other channels. and prompt injection prompt injection An attack where text the model reads, such as a web page, file or tool result, carries instructions that override the user's. It is the main security risk for agents with tools. independently. Not checked yet Adds chat.agent.sandbox.networkAccess fallback setting controlling behavior when network restrictions block a sandboxed command, defaulting to prompting a retry with unrestricted network access. Not checked yet Adds chat.agent.sandbox.allowElevation setting to prevent the agent from offering to run a command outside the sandbox when it cannot run inside it. Not checked yet Introduces Advanced Autopilot mode, which delegates turn-completion decisions to a secondary model; enabled via a dedicated preview setting. Not checked yet Adds session-scoped Allow all permission level that bypasses approvals only for the current local or Copilot CLI session without enabling global auto-approval across all workspaces. Not checked yet Adds a permission level dropdown in the chat input to switch between Manual, AI-judged, and Allow all approval modes per session. Not checked yet Introduces sandboxing sandboxing Running code inside a restricted environment that limits what it can access or change. The isolation reduces the impact of mistakes or untrusted code, depending on the sandbox's controls. as an independent security layer that restricts file system and network access for agent terminal commands regardless of permission level. Not checked yet Adds Run 'Configure Tool Approvals' from the Command Palette to review and set per-tool approval state (auto-approve run, auto-approve result, or always require manual approval) grouped by source. Not checked yet Adds Run 'Clear All Tool Approvals' from the Command Palette to reset all saved tool approvals at once. Not checked yet
14 details in Everything
VS Code Agents window gains multi-harness support with handoff, worktree isolation, and cloud PR-based task delegation︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ USE IT
Enable shell init scripts so your .bashrc profile (and activated Python environment) load before each agent command in a local Copilot session.
.vscode/settings.json
copy
"chat.agentHost.shellTool.initScript.enabled": true,
"python-envs.terminal.autoActivationType": "shellStartup"
Delegate a well-scoped coding task to the Cloud target so it runs remotely and opens a pull request — no local editor context needed.
📍 In the Agents window: select the Cloud target from the harness control › choose your GitHub repository › configure the agent role and permissions › start the session. The agent implements the task on a branch and opens a pull request for your review.
Hand off an in-progress plan to an implementation agent without losing conversation history.
📍 In the Chat view or Agents window: after the Plan agent completes its plan, select Start Implementation › choose an available implementation agent (Copilot, Claude, or Codex). VS Code carries the full conversation history and context to the new agent.
Adds a Cloud target that runs an agent harness agent harness The software around an AI agent that manages its tools, execution loop, state and limits. It determines how the model's proposed actions are run and checked. on remote infrastructure, implements a task on a branch, and opens a GitHub pull request for review.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Enables session handoff, which carries conversation history and context when switching harnesses (Copilot, Claude, Codex, or Cloud) mid-task. Not checked yet Adds code isolation via Git worktree in the Agents window, keeping agent file changes separate from the active workspace. Not checked yet Adds chat.agentHost.shellTool.initScript.enabled config key to load .bashrc / PowerShell profiles before each command in local Copilot sessions that use the SDK built-in shell tool. Not checked yet Adds python-envs.terminal.autoActivationType set to shellStartup to auto-activate the selected workspace Python environment in local Copilot agent sessions. Not checked yet
+2 more
show less
Supports running Codex on the Agent Host, making it available in the Agents window alongside the OpenAI extension Codex target. Not checked yet Adds a Start Implementation handoff action that moves from a Plan agent to an available implementation agent while preserving session context. Not checked yet
7 details in Everything
VS Code Automations let you schedule recurring agent tasks from a saved prompt, config, and schedule — no manual session start required.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ HOW TO FIND IT
Enable Automations in VS Code Stable (off by default) so the Automations panel appears in the Agents sidebar.
📍 1. Open the Settings editor (Ctrl+,).
2. Search for chat.automations.enabled.
3. Check the box to enable it.
4. Open the Agents window — the Automations entry now appears in the sidebar.
Create a daily automation that summarises the last 24 hours of Git commits and posts a grouped report — useful for async team stand-ups.
📍 1. In the Agents sidebar, select Automations › Create Automation .
2. Set Name: Daily commit summary .
3. Set Prompt: 'Summarize commits on the current branch from the last 24 hours. Group by feature, fix, and maintenance work. Include commit references and flag changes that might need documentation. Do not modify files.'
4. Choose your Git workspace and agent; optionally select New Worktree for isolation.
5. Set Schedule to Manual , select Create , then select Run now to verify output under History.
6. Once satisfied, edit the automation, set Schedule to Daily , pick a Time, ensure Enabled is checked, and select Save .
Enables chat.automations.enabled setting to unlock a new Automations panel in the Agents window sidebar for scheduling recurring agent sessions.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet Adds a new Automations feature that runs saved prompts on a schedule (Manual, Hourly, Daily, or Weekly) against a chosen workspace and agent, with run history and session review. Not checked yet Supports New Worktree isolation option for Git workspaces when creating an automation, letting the agent work on a separate branch without touching the main workspace. Not checked yet Adds a run History view per automation where each past session can be inspected, stopped, or used to diagnose failures before re-running. Not checked yet
4 details in Everything
VS Code Agents window adds BYOK BYOK Bring Your Own Key — a cloud deployment model where the customer supplies and controls the encryption keys rather than the provider, so sensitive data stays under the customer's cryptographic authority. model support, Voice Mode, quick-chat-to-workspace promotion, Automations scheduling, and worktree isolation.︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄
└──▷ HOW TO FIND IT
Use BYOK credentials (e.g. Anthropic or OpenAI) to run Agent Host sessions without signing in to GitHub.
📍 In VS Code settings, set chat.agentHost.byokModels.enabled to true, then open the Agents window. The model picker will show models from your configured BYOK providers.
Schedule a nightly code-review automation so the agent runs a saved prompt against your repo every day without manual intervention.
📍 In VS Code settings, set chat.automations.enabled to true, then open the Agents window and select Automations in the sidebar. Create a new automation, enter your saved prompt, and set the schedule to daily.
Enables chat.automations.enabled to unlock Automations — a sidebar feature for scheduling recurring agent tasks from a saved prompt, running hourly, daily, or weekly on demand. Not checked yet Adds Voice Mode to the Agents window chat input, letting practitioners steer active agent sessions by speech instead of typing. Not checked yet Adds the ability to promote a quick chat into a full workspace session, preserving title, conversation history, and the in-flight request while giving the agent access to project files. Not checked yet Adds worktree isolation mode for workspace sessions, running agent changes in an isolated Git worktree instead of directly in the working folder. Not checked yet Adds a browser preview panel scoped per session, accessible via right-click in the file explorer or a terminal link, so agents can use browser tools to inspect and interact with web applications. Not checked yet
5 details in Everything
VS Code lets you override the background utility model (commit messages, PR titles, intent detection) separately from your main chat model.
advisory CVE-2026-81380
Sep 8, 2026
·
issue 128
CVE-2026-81380: Information Disclosure through Automatic Remote Image Fetch in Chat (CVSS CVSS Common Vulnerability Scoring System, a method for describing a vulnerability's characteristics and severity. Its score is not the probability that attackers will exploit it. 5.3, affects < 1.136.2)︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet
advisory CVE-2026-81377
Sep 8, 2026
·
issue 128
CVE-2026-81377: Visual Studio Code MCP MCP Model Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. gallery metadata path traversal (CVSS CVSS Common Vulnerability Scoring System, a method for describing a vulnerability's characteristics and severity. Its score is not the probability that attackers will exploit it. 6.5, affects >= 1.0.0, < 1.136.2)︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet
advisory CVE-2026-81381
Sep 8, 2026
·
issue 128
CVE-2026-81381: Azure DevOps access token access token A credential that authorizes a client to use a service or API. It usually has limited permissions and a lifetime; anyone holding it may be able to use those permissions. disclosure through GitHub Copilot Chat workspace settings (CVSS CVSS Common Vulnerability Scoring System, a method for describing a vulnerability's characteristics and severity. Its score is not the probability that attackers will exploit it. 6.5, affects < 1.136.2)︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet
VS Code agent skills gain a forked execution context mode, isolating skill instructions from the parent agent.
Adds experimental forked skill context mode, which runs a skill in its own isolated context instead of inlining its instructions into the parent agent's context (the existing inline default behavior). Not checked yet
1 detail in Everything
code-oss gains support for nested AGENTS.md files scoped to subfolders within a workspace.
└──▷ HOW TO FIND IT
Scope AI agent instructions to a specific subfolder by placing an AGENTS.md file there instead of only at the workspace root.
📍 Place an AGENTS.md file inside a subfolder (e.g., src/backend/AGENTS.md) to apply custom instructions automatically to chat requests in that directory scope.
code-oss adds experimental multi-root workspace support for Copilot and Claude agent sessions
└──▷ USE IT
Enable multi-root workspace support for a Copilot agent session so it can reason across all folders in your workspace at once.
.vscode/settings.json
copy
{
"chat.agentHost.copilotAgent.multiRootEnabled": true
}
Enable multi-root workspace support for a Claude agent session to let it operate across every folder in a multi-root workspace.
.vscode/settings.json
copy
{
"chat.agentHost.claudeAgent.multiRootEnabled": true
}
Adds chat.agentHost.copilotAgent.multiRootEnabled setting to enable experimental multi-root workspace support for Copilot agent sessions in the editor window. Not checked yet Adds chat.agentHost.claudeAgent.multiRootEnabled setting to enable experimental multi-root workspace support for Claude agent sessions in the editor window. Not checked yet Agent hooks remain scoped to a single workspace folder; when hooks exist in multiple folders, VS Code prompts the user to select the primary folder from which to load them. Not checked yet
3 details in Everything
VS Code Agents window adds Agent Merge to autonomously resolve PR blockers and merge pull requests hands-free.
└──▷ HOW TO FIND IT
Enable Agent Merge so VS Code autonomously resolves PR review comments, failed CI checks, and merge conflicts, then merges the PR without manual intervention.
📍 In the Agents window, set chat.agentMerge.enabled to true, open a session associated with a pull request, then select Agent Merge in the title bar › Enable Agent Merge , and configure which blockers to address and whether to auto-merge.
Stop Agent Merge from monitoring a pull request mid-session if you want to resume manual control.
📍 In the Agents window title bar, select Agent Merge , then select Disable Agent Merge .
New Agents window provides a dedicated VS Code interface for running and tracking multiple agent sessions in parallel across workspaces. Not checked yet Enables chat.agentMerge.enabled setting to activate Agent Merge, which monitors a PR and autonomously addresses unresolved review threads, failed required CI checks, branch conflicts, and merges or queues the PR when ready. Not checked yet Supports attaching folders, repositories, GitHub issues, and pull requests as context before sending the first prompt in an agent session. Not checked yet Supports remote workspace connections via SSH or dev tunnel through the Remote Setup control in the Agents window. Not checked yet Adds chat personalization options including decorative chat backgrounds and an interactive VS Code pet. Not checked yet
5 details in Everything
VS Code introduces the Agent Host — a persistent, portable process for agent sessions — plus the open Agent Host Protocol (AHP AHP Analytic Hierarchy Process, a structured decision-making method that breaks a complex choice into a weighted hierarchy of criteria and scores alternatives against them, letting cyber tools rank or prioritize options mathematically. ) and multi-language client libraries.
└──▷ TRY IT
Start a standalone Agent Host so external AHP clients can connect to and monitor agent sessions running on your machine.
Experience a persistent session that survives closing its folder: start a session, close the folder, then reopen it or switch to the Agents window to resume.
📍 1. Open a folder in an editor window, select Copilot from the harness picker, and start an agent session.
2. While the agent is working, close the folder (do not quit VS Code).
3. Reopen the folder — session state and progress are preserved.
4. Open the Agents window to monitor the same live session from both surfaces simultaneously.
Connect to a remote Agent Host from the browser to review and manage a session running on another machine.
📍 1. On the remote machine, start VS Code with your workspace and begin an agent session.
2. In the Agents window, connect to the remote Agent Host (configured via SSH or dev tunnel).
3. Navigate to insiders.vscode.dev/agents in a browser, select the same host, and continue the session.
Adds the code agent host (or code-insiders agent host) subcommand to start a standalone Agent Host process that keeps agent sessions alive independent of any editor window. Not checked yet Introduces the Agent Host Protocol (AHP), an open, URI-addressable protocol for coordinating multiple clients around the same long-running agent session, with channels for sessions, chats, terminals, and changesets. Not checked yet Ships AHP client libraries in Rust (ahp, ahp-types, ahp-ws), TypeScript (@microsoft/agent-host-protocol), Kotlin (com.microsoft.agenthostprotocol:agent-host-protocol), Go (github.com/microsoft/agent-host-protocol/clients/go), and Swift (Swift Package Manager: microsoft/agent-host-protocol). Not checked yet Enables agent sessions to persist after closing a folder or editor window, with state and progress resuming when the folder is reopened. Not checked yet Enables remote agent sessions: the Agent Host can run on a remote machine (via SSH, dev tunnels, or the web at insiders.vscode.dev/agents) while clients connect from the desktop or browser. Not checked yet
+2 more
show less
Adds the Agents window as a synchronized client surface, letting practitioners monitor and interact with a live agent session from both the editor window and the Agents window without creating a copy. Not checked yet Supports multiple agent harnesses agent harnesses The software around an AI agent that manages its tools, execution loop, state and limits. It determines how the model's proposed actions are run and checked. (Copilot via GitHub Copilot SDK, Claude via Anthropic Claude Agent SDK) under a common session model while each harness retains its own agent loop, tools, and provider-specific capabilities. Not checked yet
7 details in Everything
VS Code adds experimental migration flow to move user customizations to agent-host folders via chat.customizations.userDataMigration.enabled
Adds experimental Migrate User Data Customizations flow, enabled via the chat.customizations.userDataMigration.enabled setting, to move user-level agents and instructions from VS Code profile user data into harness-agnostic folders readable by the active agent host. Not checked yet Migration UI appears in the Agent Customizations editor (opened via Configure Chat gear icon) under the active agent host, letting users select individual agents and instructions — with per-file preview — before committing; a confirmation step offers optional deletion of the originals from profile user data. Not checked yet
2 details in Everything
code-oss gains multi-group chat layouts, side chats, and cross-app session discovery with chat.agentSessions.showExternal
└──▷ USE IT
Show only external sessions updated in the last 24 hours to quickly pick up where you left off in Claude Code or Copilot CLI.
.vscode/settings.json
copy
"chat.agentSessions.showExternal": "Last 24 Hours"
Split the current chat group to the right so you can run a side investigation without polluting your main conversation.
📍 Sessions: Split Chat Group Right
Adds chat.agentSessions.showExternal setting to control visibility of external sessions (values: None , Recent , Last 24 Hours , Last 7 Days , All ); defaults to hiding external sessions. Not checked yet Adds Sessions: Split Chat Group Right, Sessions: Split Chat Group Down, Sessions: Move Chat to Previous Group, and Sessions: Move Chat to Next Group Command Palette commands for keyboard-driven chat layout management. Not checked yet Supports multi-group chat layouts in agent host sessions — drag chat tabs to edges to split into side-by-side or stacked groups, with focus navigation via Ctrl+K Ctrl+Left / Ctrl+K Ctrl+Right (macOS: ⌘K ⌘← / ⌘K ⌘→). Not checked yet Introduces side chats that privately inherit the source conversation as context and open in a group beside the source chat without polluting the main conversation history. Not checked yet Enables discovery and continuation of local agent sessions created outside VS Code in Copilot CLI, the GitHub Copilot app, Claude Code, and Codex from the Chat view or Agents window. Not checked yet
+1 more
show less
For Copilot sessions, external session discovery includes repository-associated sessions updated within the last seven days. Not checked yet
6 details in Everything
VS Code adds ChatGPT subscription support and experimental Agent Host for Codex, with dual-provider model picking in Chat view.
Adds ChatGPT subscription as a second authentication option for Codex on the Agent Host — sign in via the account menu (Sign in to ChatGPT ) using a free ChatGPT account, no Copilot Pro+ required. Not checked yet Adds dual-provider model picker in the Chat view that groups models under ChatGPT and Copilot when both accounts are signed in, identifying same-named models by provider and persisting the chosen provider per session. Not checked yet Adds experimental Use Codex without GitHub sign-in option so ChatGPT-signed-in users can run ChatGPT-backed Codex models in the desktop Agents window without a GitHub account. Not checked yet Adds experimental Agent Host implementation for running Codex, available as an opt-in alongside the default Chat view path. Not checked yet
4 details in Everything
VS Code Agents window gains ChatGPT/Codex sign-in, Anthropic credential support, and multi-chat layout within a single session
└──▷ USE IT
Use a personal ChatGPT subscription to back Codex in the Agents window without signing in to GitHub.
.vscode/settings.json
copy
// In VS Code settings, enable the setting, then open the Agents window account menu and choose 'Sign in to ChatGPT'
"chat.agentHost.codexAgent.enabled": true
Enables chat.agentHost.codexAgent.enabled to use a ChatGPT subscription as the Codex backend — open the account menu in the Agents window and select Sign in to ChatGPT to authenticate without requiring a GitHub sign-in. Not checked yet Supports opening the Agents window on desktop without GitHub sign-in when Claude is configured with Anthropic credentials or Codex is already signed in to ChatGPT. Not checked yet Model picker now filters to only providers with available credentials when signed out of GitHub, and disambiguates by provider name when Copilot and ChatGPT both offer a model with the same name. Not checked yet Shows a notification when VS Code discovers an existing Claude configuration, indicating Claude is available without GitHub sign-in. Not checked yet Supports multiple independent peer chats, side chats, and read-only subagent chats within a single agent host session, arrangeable in horizontal or vertical groups. Not checked yet
5 details in Everything
VS Code adds full MCP MCP Model Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. server support: gallery install, mcp.json config, CLI provisioning, sandboxing sandboxing Running code inside a restricted environment that limits what it can access or change. The isolation reduces the impact of mistakes or untrusted code, depending on the sandbox's controls. , and dev container integration.
└──▷ TRY IT
Provision a fetch MCP server into your user profile from the terminal so it is available across all workspaces immediately.
copy
$ code --add-mcp "{\"name\":\"mcp-fetch\",\"command\":\"uvx\",\"args\":[\"mcp-server-fetch\"]}"
Sandbox a local stdio MCP server so it can only write inside the workspace folder and reach one external API domain, reducing blast radius if the server is compromised.
json
copy
{
"servers": {
"myServer": {
"type": "stdio",
"command": "npx",
"args": ["-y", "@example/mcp-server"],
"sandboxEnabled": true
}
},
"sandbox": {
"filesystem": {
"allowWrite": ["${workspaceFolder}"]
},
"network": {
"allowedDomains": ["api.example.com"]
}
}
}
Share a remote and a local MCP server configuration with your whole team by committing .vscode/mcp.json to source control.
json
copy
{
"servers": {
"github": {
"type": "http",
"url": "https://api.githubcopilot.com/mcp"
},
"playwright": {
"command": "npx",
"args": ["-y", "@microsoft/mcp-server-playwright"]
}
}
}
Adds MCP server gallery browsable via @mcp in the Extensions view, with Install for user profile or Install in Workspace to write to .vscode/mcp.json. Not checked yet Adds .vscode/mcp.json (workspace) and user-profile mcp.json (opened via MCP: Open User Configuration) for manually configuring stdio and HTTP MCP servers under a servers key. Not checked yet Adds --add-mcp CLI flag to provision MCP servers into the user profile from the command line, accepting a JSON server config as {"name":"server-name","command":...}. Not checked yet Adds sandboxEnabled: true field in mcp.json server entries and a top-level sandbox object with filesystem.allowWrite and network.allowedDomains rules to isolate stdio MCP servers on macOS and Linux. Not checked yet Adds chat.mcp.discovery.enabled setting to automatically detect and reuse MCP server configurations from other applications such as Claude Desktop. Not checked yet
+10 more
show less
Adds chat.mcp.autostart setting Experimental to automatically restart MCP servers when their configuration changes. Not checked yet Adds customizations.vscode.mcp.servers section in devcontainer.json for including MCP server configurations in Dev Containers Dev Containers A Docker-based development environment specification, maintained by Microsoft and the open-source community, that lets a tool ship a fully configured, reproducible workspace any developer can launch inside VS Code or a compatible editor. . Not checked yet Supports a portable ~/.copilot/mcp-config.json file and workspace-level .mcp.json that the Agent Host reads natively, in addition to .vscode/mcp.json. Not checked yet Adds MCP: Open Remote User Configuration command to define MCP servers that run on a remote machine when connected via Remote Development. Not checked yet Enables per-server enable/disable state (stored separately from mcp.json) via right-click in the MCP SERVERS - INSTALLED Extensions view section or MCP: List Servers Command Palette action. Not checked yet Adds MCP server configuration synchronization across devices via Settings Sync, enabled through Settings Sync: Configure with the MCP Servers option. Not checked yet Exposes MCP Resources as chat context via Add Context > MCP Resources or the MCP: Browse Resources command. Not checked yet Supports MCP prompt templates invocable via /<MCP server>.<prompt> syntax in the chat input. Not checked yet Supports MCP Apps — interactive UI components (forms, visualizations, drag-and-drop lists) rendered inline in chat. Not checked yet Adds Chat: Open Customizations Command Palette command to open the Agent Customizations editor Preview for discovering and managing MCP servers and other agent customizations. Not checked yet
15 details in Everything
VS Code adds a Language Models editor with BYOK BYOK Bring Your Own Key — a cloud deployment model where the customer supplies and controls the encryption keys rather than the provider, so sensitive data stays under the customer's cryptographic authority. support, model pinning, thinking-effort controls, and a Custom Endpoint provider for self-hosted APIs.
└──▷ USE IT
Connect a self-hosted Anthropic-compatible endpoint as a BYOK model without a Copilot plan, using the Messages API type.
json
copy
// chatLanguageModels.json
[
{
"name": "Anthropic",
"vendor": "customendpoint",
"apiKey": "${input:anthropicApiKey}",
"apiType": "messages",
"models": [
{
"id": "claude-sonnet-4-6",
"name": "Claude Sonnet 4.6",
"toolCalling": true,
"vision": true
}
]
}
]
Enable BYOK models inside Copilot Agent Host sessions (e.g., the Agents window) so your custom-endpoint models are available there too.
json
copy
// settings.json
{
"chat.agentHost.byokModels.enabled": true
}
Register an Azure OpenAI deployment using Entra ID Entra ID Microsoft's cloud-based identity and access management service, formerly Azure Active Directory, used by cyber tools to authenticate users and enforce access policies in Microsoft 365 and Azure environments. auth with a large context window context window The maximum amount of text, counted in tokens, a model can take in and produce in one request. Everything the model knows about the task has to fit inside it. for complex agentic tasks.
json
copy
// chatLanguageModels.json
[
{
"name": "Azure",
"vendor": "azure",
"models": [
{
"id": "<my-deployment-name>",
"name": "GPT-5.5",
"url": "https://<my-endpoint>.openai.azure.com",
"toolCalling": true,
"vision": true,
"maxInputTokens": 200000,
"maxOutputTokens": 64000
}
]
}
]
New Chat: Manage Language Models command and gear-icon entry point opens a Language Models editor where you can view, filter, and manage all available models. Not checked yet Model search and filtering in the Language Models editor supports text search plus @provider:, @capability: (values: tools, vision, agent), and @visible:true/false filter syntax. Not checked yet New chatLanguageModels.json config file stores built-in-provider and custom-endpoint model configuration (API key, endpoint URL, apiType, toolCalling, vision, maxInputTokens, maxOutputTokens, etc.). Not checked yet New chat.agentHost.byokModels.enabled setting Experimental enables BYOK models in Agent Host / Copilot Agents-window sessions; takes effect after the agent host process is restarted. Not checked yet New Custom Endpoint provider (replaces the deprecated OpenAI Compatible provider) supports three apiType values — Chat Completions , Responses , and Messages (Anthropic) — configurable per provider or per model in chatLanguageModels.json. Not checked yet
+9 more
show less
Configurable thinking-effort submenu in the model picker lets you set effort level per reasoning model; the picker label reflects the selection (e.g., Claude Sonnet 4.6 · High ) and the choice is persisted per session. Not checked yet Pin favorite models to a dedicated Pinned section at the top of the model picker via a pin icon on hover. Not checked yet Model visibility (shown/hidden in picker) is now controlled per-model via an eye icon in the Language Models editor. Not checked yet Bring Your Own Key (BYOK) support lets you connect to Azure, Anthropic, Gemini, OpenAI, custom endpoints, or local models (e.g., Ollama via the official Ollama extension) without a GitHub account or Copilot plan. Not checked yet Built-in Ollama provider is deprecated; the official Ollama extension (from the Ollama publisher on the Marketplace) is now the recommended path for local Ollama BYOK models. Not checked yet Auto model selection (select Auto in the model picker) routes each request to the optimal model based on task complexity and real-time model availability; hover over a chat response to see which model was used. Not checked yet The github.copilot.chat.anthropic.thinking.effort and github.copilot.chat.responsesApiReasoningEffort settings are deprecated in favor of the new in-picker thinking-effort submenu. Not checked yet The github.copilot.chat.customOAIModels setting is deprecated in favor of the new Custom Endpoint provider and chatLanguageModels.json. Not checked yet Extensions providing language model providers are now discoverable in the Extensions view via the @tag:language-models filter. Not checked yet
14 details in Everything
└──▷ BREAKING ON UPGRADE
! The github.copilot.chat.anthropic.thinking.effort and github.copilot.chat.responsesApiReasoningEffort settings are deprecated; thinking effort must now be configured from the language model picker UI.! The github.copilot.chat.customOAIModels setting is deprecated; custom endpoint models must now be configured via the Custom Endpoint provider in chatLanguageModels.json.! The built-in Ollama provider is deprecated; users relying on it for BYOK local models must install the official Ollama extension and remove the built-in provider configuration to continue using Ollama models.
VS Code gains custom agents via .agent.md files with handoffs, subagents, hooks, and per-agent tool/model control
└──▷ USE IT
Create a read-only planning agent that hands off to an implementation agent, preventing accidental code edits during the planning phase.
yaml
copy
---
name: Planner
description: Generate an implementation plan for new features or refactoring existing code.
tools: ['web/fetch', 'search/codebase', 'search/usages']
model: ['Claude Opus 4.5', 'GPT-5.2']
handoffs:
- label: Implement Plan
agent: implementer
prompt: Implement the plan outlined above.
send: false
---
You are in planning mode. Generate a detailed implementation plan. Do not make any code edits.
Create a coordinating Feature Builder agent that restricts subagent invocations to named specialist agents for a research-then-implement workflow.
yaml
copy
---
name: Feature Builder
description: Build features by researching first, then implementing
tools: ['agent']
agents: ['Researcher', 'Implementer']
---
You are a feature builder. For each task:
1. Use the Researcher agent to gather context and find relevant patterns in the codebase
2. Use the Implementer agent to make the actual code changes based on research findings
Open the Agent Customizations editor to browse, create, and manage all custom agents in one place.
📍 1. Open the Command Palette (Ctrl+Shift+P / Cmd+Shift+P).
2. Run Chat: Open Customizations .
3. In the Agent Customizations editor Preview , select New Custom Agent to create an agent, or select an existing agent to edit its .agent.md file.
Adds custom agents defined in .agent.md Markdown files, stored in .github/agents (workspace), .claude/agents (Claude format), or ~/.copilot/agents (user profile) to give chat personas their own instructions, tools, and model selections. Not checked yet Introduces handoffs frontmatter field in .agent.md to define handoff buttons that transition between agents after a response, with handoffs.label, handoffs.agent, handoffs.prompt, handoffs.send (auto-submit, default false), and handoffs.model (e.g. GPT-5 (copilot) or Claude Sonnet 4.5 (copilot)) subfields. Not checked yet Adds tools frontmatter field to restrict which tools (built-in, MCP MCP Model Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. , extension-contributed, or tool sets) are available per agent; use <server name>/* to include all tools from an MCP server. Not checked yet Adds agents frontmatter field to control which agents are available as subagents; use * to allow all, [] to disallow all, or a named list; requires agent in tools. Not checked yet Adds model frontmatter field to pin an agent to a specific model (string) or a prioritized fallback list (array) tried in order until an available model is found. Not checked yet
+13 more
show less
Adds user-invocable frontmatter flag (boolean, default true) to hide an agent from the chat dropdown while still allowing subagent invocation. Not checked yet Adds disable-model-invocation frontmatter flag (boolean, default false) to prevent an agent from being invoked as a subagent by other agents. Not checked yet Adds hooks Preview frontmatter field to scope hook commands to a specific agent, using the same format as hook configuration files; requires chat.useCustomAgentHooks setting. Not checked yet Adds argument-hint frontmatter field to display hint text in the chat input field guiding users on how to interact with the agent. Not checked yet Adds description and name frontmatter fields to set the agent's display name and placeholder description in the chat UI. Not checked yet Adds mcp-servers frontmatter field to supply MCP server config JSON for agents targeting github-copilot (target: github-copilot). Not checked yet Supports #tool:<tool-name> syntax (e.g. #tool:web/fetch) in the agent file body to reference specific tools in the agent's instructions. Not checked yet Adds chat.agentFilesLocations setting to configure additional file locations for workspace custom agent files. Not checked yet Adds chat.useCustomizationsInParentRepositories setting to discover custom agents from a parent repository root in monorepos. Not checked yet Adds chat.subagents.allowInvocationsFromSubagents setting to allow self-referential agents that list themselves in the agents property. Not checked yet Adds Agent Customizations editor Preview accessible via Chat: Open Customizations from the Command Palette to discover, create, and manage agent customizations. Not checked yet Adds Chat: New Custom Agent command to create user-level custom agents directly from the Command Palette. Not checked yet Deprecates infer frontmatter field in favor of the independent user-invocable and disable-model-invocation fields. Not checked yet
18 details in Everything
└──▷ BREAKING ON UPGRADE
! When Agent Host is enabled, user-level custom agents are read from ~/.copilot/agents instead of VS Code profile user data — existing user-level agents stored elsewhere must be moved to ~/.copilot/agents to continue working.! The infer frontmatter field is deprecated; infer: false behavior (hiding from picker and blocking subagent invocation) must now be expressed with the separate user-invocable: false and disable-model-invocation: true fields.
VS Code gains Agent Skills: portable, on-demand capability bundles for GitHub Copilot loaded via SKILL.md files
└──▷ USE IT
Create a project skill for Playwright testing so Copilot automatically loads test-running instructions whenever you ask about browser tests.
yaml
copy
# .github/skills/webapp-testing/SKILL.md
---
name: webapp-testing
description: Guide for testing web applications using Playwright. Use this when asked to create or run browser-based tests.
---
# Web Application Testing with Playwright
Use Playwright locators and the [test template](./test-template.js) to create atomic, role-based tests in the `tests/` directory.
Run a large PR-review skill in a forked subagent context so its intermediate reasoning does not pollute the main conversation window.
yaml
copy
# .github/skills/review-pr/SKILL.md
---
name: review-pr
description: Review a pull request for code quality, style, and correctness. Use when asked to review a PR.
context: fork
---
# PR Review
Follow these steps to review the pull request...
Quickly scaffold a new skill with AI rather than writing SKILL.md by hand.
copy
$ /create-skill a skill for running and debugging integration tests
Adds Agent Skills support, letting practitioners define reusable capability bundles in SKILL.md files that GitHub Copilot loads on-demand across VS Code, Copilot CLI, and the Copilot cloud agent. Not checked yet Supports project skills stored in .github/skills/, .claude/skills/, or .agents/skills/, and personal skills in ~/.copilot/skills/, ~/.claude/skills/, or ~/.agents/skills/. Not checked yet Adds chat.agentSkillsLocations setting to configure additional project skill directories beyond the defaults. Not checked yet Adds chat.useCustomizationsInParentRepositories setting to enable skill discovery from the parent repository root in monorepos. Not checked yet Adds SKILL.md frontmatter fields name, description, argument-hint, user-invocable, disable-model-invocation, and context to control skill metadata, slash-command visibility, auto-loading behavior, and execution mode. Not checked yet
+5 more
show less
Adds context: fork frontmatter value to run a skill in a dedicated subagent context, returning only the final result to the parent agent's conversation window — requires the github.copilot.chat.skillTool.enabled setting. Not checked yet Adds /create-skill slash command in chat to AI-generate a SKILL.md file and directory structure from a natural-language description. Not checked yet Adds /skills slash command to quickly open the Configure Skills menu from the chat input. Not checked yet Exposes skills as slash commands in the chat / menu, with user-invocable and disable-model-invocation frontmatter properties controlling whether each skill appears in the menu and whether Copilot auto-loads it. Not checked yet Adds an Agent Customizations editor Preview — opened via Chat: Open Customizations from the Command Palette — with a Skills tab for discovering, creating, and managing skills including a Generate Skill AI-generation option. Not checked yet
10 details in Everything
VS Code adds structured custom instructions via .github/copilot-instructions.md, AGENTS.md, CLAUDE.md, and *.instructions.md files to shape AI chat responses.
└──▷ USE IT
Enforce workspace-wide coding standards automatically on every chat request without repeating them in each prompt.
markdown
copy
# .github/copilot-instructions.md
# Project general coding standards
## Naming Conventions
- Use PascalCase for component names, interfaces, and type aliases
- Use camelCase for variables, functions, and methods
- Use ALL_CAPS for constants
## Error Handling
- Use try/catch blocks for async operations
- Always log errors with contextual information
Apply language-specific rules only to Python files, keeping general standards separate from per-language conventions.
markdown
copy
---
name: 'Python Standards'
description: 'Coding conventions for Python files'
applyTo: '**/*.py'
---
# Python coding standards
- Follow the PEP 8 style guide.
- Use type hints for all function signatures.
- Write docstrings for public functions.
Lock a workspace to only workspace-level instruction sources and disable user-profile instructions for reproducible team behavior.
json
copy
"chat.instructionsFilesLocations": {
".github/instructions": true,
".claude/rules": true,
"~/.copilot/instructions": false,
"~/.claude/rules": false
}
Adds .github/copilot-instructions.md as a workspace-wide always-on instruction file automatically applied to all chat requests. Not checked yet Adds AGENTS.md file support for always-on instructions, useful when working with multiple AI agents; can be placed in the workspace root or subfolders (experimental subfolder scoping). Not checked yet Adds CLAUDE.md file support for compatibility with Claude Code and other Claude-based tools, stored in the workspace root, .claude folder, or user home directory. Not checked yet Adds *.instructions.md file-based instructions that apply conditionally via applyTo glob patterns (e.g. **/*.py, **/*.ts,**/*.tsx) or semantic task matching. Not checked yet Adds chat.instructionsFilesLocations setting to configure which folders VS Code searches for instructions files, supporting per-scope enable/disable (e.g. disable ~/.copilot/instructions for workspace-only policies). Not checked yet
+11 more
show less
Adds chat.useCustomizationsInParentRepositories setting to discover instructions files from a parent repository root in monorepos. Not checked yet Supports ~/.copilot/instructions and ~/.claude/rules as user-level instruction folders read by the Agent Host, independent of VS Code profile data. Not checked yet Adds YAML frontmatter header fields name, description, and applyTo to .instructions.md files to control display name, hover description, and automatic file-pattern targeting. Not checked yet Supports #tool:<tool-name> syntax inside instruction file bodies to reference agent tools (e.g. #tool:web/fetch). Not checked yet Adds /create-instruction chat command to AI-generate a targeted .instructions.md file from a natural-language description of a convention. Not checked yet Adds /instructions shortcut in the chat input to quickly open the Configure Instructions and Rules menu. Not checked yet Adds Chat: New Instructions File Command Palette command to create workspace or user-scoped instructions files. Not checked yet Adds Chat: Open Customizations Command Palette command to open the Agent Customizations editor Preview for discovering, creating, and managing all agent customizations. Not checked yet Supports recursive subfolder organization inside instruction file locations (e.g. .github/instructions/frontend/react.instructions.md). Not checked yet Supports Markdown links inside instruction files to reference specific context such as other instruction files or URLs. Not checked yet Allows extracting project conventions from an ongoing chat session by prompting the agent to 'extract an instruction from this'. Not checked yet
16 details in Everything
└──▷ BREAKING ON UPGRADE
! When Agent Host is enabled, user-level instructions are read from ~/.copilot/instructions and ~/.claude/rules rather than VS Code profile user data — existing user-level instructions stored elsewhere will not be picked up.
VS Code gains an Agent Customizations editor, monorepo discovery setting, and a preview evaluation extension for AI customization files.
└──▷ USE IT
Enable monorepo customization discovery so that shared instructions and prompt files at the repo root apply when you open a subfolder like packages/frontend/ in VS Code.
json
copy
// In VS Code settings (settings.json)
{
"chat.useCustomizationsInParentRepositories": true
}
Analyze an instructions file for contradictions, ambiguous wording, and cross-file conflicts before relying on it in production.
📍 1. Open the customization file (e.g. style.instructions.md) in the editor.
2. Open the Command Palette (Ctrl+Shift+P / Cmd+Shift+P).
3. Run Chat Customizations Evaluations: Analyze.
4. Review diagnostics in the Problems panel (Ctrl+Shift+M / Cmd+Shift+M).
5. Select Implement Suggestions to apply recommended improvements.
Run the Waza evaluation suite against a skill file to measure how well it performs against a defined set of test cases.
📍 1. Run Chat Customizations Evaluations: Download Waza Binary from the Command Palette to install Waza.
2. Open your SKILL.md file, then run Chat Customizations Evaluations: Create Waza Eval Scaffold to generate evaluation files.
3. Run Chat Customizations Evaluations: Run Waza Evaluation to execute the test suite and review results.
Adds the Agent Customizations editor — a central UI with per-type tabs, syntax highlighting, and validation — opened via Chat: Open Customizations from the Command Palette or the Customizations panel in the left sidebar. Not checked yet Adds Chat: New <customization-type> Command Palette command to scaffold a new customization file at the correct location for the selected type. Not checked yet Adds the chat.useCustomizationsInParentRepositories setting (disabled by default) to make VS Code walk up the folder hierarchy to the nearest .git root and discover customizations — including copilot-instructions.md, AGENTS.md, CLAUDE.md, *.instructions.md, *.prompt.md, *.agent.md, and hooks — enabling monorepo setups where only a subfolder is opened. Not checked yet Adds the Chat Customizations Evaluations extension (preview, installed separately from the Visual Studio Marketplace) to analyze SKILL.md, *.agent.md, *.instructions.md, and *.prompt.md files for contradictions, ambiguous wording, persona conflicts, cognitive load, and cross-file conflicts; diagnostics surface in the Problems panel with line/column locations. Not checked yet Adds the /analyze-prompt slash command in the Chat view to run customization file analysis and surface diagnostics inline. Not checked yet
+5 more
show less
Adds Chat Customizations Evaluations: Analyze Command Palette command to trigger analysis of the active customization file. Not checked yet Adds Chat Customizations Evaluations: Run Waza Evaluation (plus Chat Customizations Evaluations: Download Waza Binary and Chat Customizations Evaluations: Create Waza Eval Scaffold) to measure skill file performance against test cases using the Waza evaluation framework. Not checked yet Adds the Developer: Open Agent Debug Panel Command Palette command to open the Agent Debug Logs panel for troubleshooting customization issues. Not checked yet Supports AI-assisted customization creation: entering a natural-language prompt in the Overview tab of the Agent Customizations editor generates a correctly formatted customization file. Not checked yet When Agent Host is enabled, user-level customizations are read from harness-agnostic folders (~/.copilot for Copilot, ~/.claude for Claude) instead of the VS Code profile user data directory. Not checked yet
10 details in Everything
VS Code Copilot gains searchable session history sync, Chronicle query commands, and per-repo exclusion controls.
└──▷ USE IT
Prevent a sensitive private repo from ever syncing session data to the cloud.
json
copy
// .vscode/settings.json
{
"chat.sessionSync.excludeRepositories": [
"my-org/private-repo",
"my-org/secret-*"
]
}
Generate a standup report from the last 24 hours of Copilot sessions, grouped by branch and repo.
copy
$ /chronicle:standup
Find all sessions where you worked on the payments API to resume context quickly.
copy
$ /chronicle:search payments API
Adds chat.sessionSync.enabled setting (default true) to sync local Copilot agent sessions to your GitHub account across devices; set to false to keep session data local only. Not checked yet Adds chat.sessionSync.excludeRepositories setting (default []) accepting exact owner/repo names or glob patterns (e.g. my-org/secret-*) to prevent specific repositories from syncing to the cloud. Not checked yet Adds github.copilot.chat.localIndex.enabled setting (default true) as the prerequisite toggle for local session tracking. Not checked yet Adds Delete Session Sync Data command (github.copilot.sessionSync.deleteSessions) in the Command Palette with a picker to delete sessions from local and cloud, or from cloud only. Not checked yet Adds Reindex Sessions command (github.copilot.chronicle.reindex) in the Command Palette to rebuild the local session index. Not checked yet
+10 more
show less
Adds /chronicle:standup chat command to summarize recent coding sessions into a standup report grouped by branch and repository, covering the last 24 hours by default. Not checked yet Adds /chronicle:tips chat command to analyze recent session history and suggest more effective Copilot usage patterns. Not checked yet Adds /chronicle:cost-tips chat command to identify opportunities to reduce token usage and Copilot cost. Not checked yet Adds /chronicle:search <query> chat command to search sessions by keyword, file path, or pull request/issue reference. Not checked yet Adds /chronicle:reindex chat command to rebuild the local session index and sync session data. Not checked yet Adds Chronicle free-form natural-language querying over session history (e.g. 'What files did I edit yesterday?') using semantic search semantic search Searching by meaning rather than requiring the same words. It commonly compares vector representations to find related content, even when the wording differs. across sessions. Not checked yet Session history is stored in a local SQLite database, tracking metadata (repository, branch, timestamps, agent), conversation turns, files touched, and external references (PRs, issues, commit SHAs); secrets are filtered before sync. Not checked yet Adds enterprise policy controls: a GitHub.com policy (Store local sessions in the Cloud ) and a VS Code group policy (CopilotSessionSync) that forces chat.sessionSync.enabled to false when disabled, showing Disabled by policy status. Not checked yet Adds session sync status indicator in the Copilot status bar in the Chat view, showing states including Not enabled , Enabled , N sessions synced , Syncing N sessions , Disabled by policy , and Sync error . Not checked yet Adds session sharing via the Agents tab on GitHub.com, providing view-only access to repository collaborators for prompts, responses, and file changes; shared sessions are not indexed for other users' queries. Not checked yet
15 details in Everything
VS Code adds full agent session management: multi-chat tabs, context compaction context compaction Condensing an AI conversation or working history to free space in the model's context window. A summary preserves useful information, but details can be lost. , side chats, and a cross-project sessions list.
└──▷ USE IT
Prevent VS Code from automatically compacting long agent conversations so you retain full history.
json
copy
// settings.json
{
"github.copilot.chat.summarizeAgentConversationHistory.enabled": false
}
Compact a long session while preserving specific decisions, to reduce token cost on the next request.
copy
$ /compact focus on the database schema decisions
Ask a clarifying question mid-task without polluting the main chat thread.
copy
$ /btw why did the agent choose a LEFT JOIN here instead of an INNER JOIN?
Adds github.copilot.chat.summarizeAgentConversationHistory.enabled setting (default true) to control automatic context compaction; set to false to disable. Not checked yet Adds /compact chat command to manually summarize earlier conversation history, with optional inline instructions (e.g. /compact focus on the database schema decisions). Not checked yet Adds /btw <question> chat command to open a side chat that privately inherits current conversation context without polluting the main chat transcript. Not checked yet Adds chat.viewSessions.enabled setting to show or hide the sessions list in the Chat view. Not checked yet Adds ⌘T / Ctrl+T keyboard shortcut to create a new chat tab within an existing session, and ⇧⌘T / Ctrl+Shift+T to reopen the most recently closed chat. Not checked yet
+5 more
show less
Adds multi-chat tabs within a single agent host session: each chat has its own conversation, title, agent, and model selection, but shares the session's workspace and worktree. Not checked yet Adds side chats (Copilot and Claude sessions in the Agents window only) that branch from a turn — including in-progress responses — and favor explanation over action. Not checked yet Adds a Session Target control in both the Chat view and Agents window for choosing an agent harness agent harness The software around an AI agent that manages its tools, execution loop, state and limits. It determines how the model's proposed actions are run and checked. (Local, Copilot, Claude, Codex, or Cloud) per session. Not checked yet Adds an Agents window with a cross-workspace sessions list that supports custom groups, drag-and-drop reordering, pinning, archiving, and grouping by workspace or timeframe. Not checked yet Adds per-session configuration for Agent role (Agent, Plan, Ask), Permission level (tool-approval autonomy), and Language model , with the Agents window remembering the last-used Agent and Permission level as defaults for new sessions. Not checked yet
10 details in Everything
VS Code gains subagent support — delegate subtasks to isolated AI agents with parallel execution, custom agent routing, and model selection.
└──▷ USE IT
Restrict a TDD coordinator agent to only the Red, Green, and Refactor subagents so it never accidentally delegates to a generic coding agent.
yaml
copy
---
name: TDD
tools: ['agent']
agents: ['Red', 'Green', 'Refactor']
---
Implement the following feature using test-driven development. Use subagents to guide the following steps:
1. Use the Red agent to write failing tests
2. Use the Green agent to implement code to pass the tests
3. Use the Refactor agent to improve the code quality
Create an internal helper agent that is invisible in the chat dropdown and can only be invoked by other agents as a subagent.
yaml
copy
---
name: internal-helper
user-invocable: false
---
This agent can only be invoked as a subagent.
Invoke a subagent from a prompt file to do isolated research before editing docs, using the agent tool in the tools frontmatter.
yaml
copy
---
name: document-feature
tools: ['agent', 'read', 'search', 'edit']
---
Run a subagent to research the new feature implementation details and return only information relevant for user documentation.
Then update the docs/ folder with the new documentation.
Adds the agent/runSubagent tool that the main agent uses to delegate subtasks to isolated subagents; must be enabled for subagent invocation to work. Not checked yet Adds chat.subagents.allowInvocationsFromSubagents setting to enable recursive nested subagent spawning (off by default to prevent infinite loops). Not checked yet Adds chat.subagents.useRichRendering setting to control whether subagents open in their own editor (default) or display activity inline in the parent chat. Not checked yet Adds user-invocable frontmatter property to custom agent .agent.md files — set to false to hide an agent from the dropdown and make it only accessible as a subagent. Not checked yet Adds disable-model-invocation frontmatter property to custom agent .agent.md files — set to true to prevent an agent from being invoked as a subagent by other agents. Not checked yet
+8 more
show less
Adds agents frontmatter property to custom agent .agent.md files to restrict which agents a coordinator may delegate to; accepts a list of names (e.g. ['Red', 'Green', 'Refactor']), * for all, or [] to block subagent use entirely. Not checked yet Adds model frontmatter property to custom agent .agent.md files — accepts a single model name or a prioritized list — to pin a subagent to a specific model; cannot exceed the cost tier of the main model. Not checked yet Supports including 'agent' (or runSubagent) in a prompt file's tools frontmatter array to invoke subagents from .prompt.md files. Not checked yet Subagents run statelessly and in parallel when the main agent is prompted with tasks structured for isolated or concurrent analysis. Not checked yet Each subagent appears in the Chat view as a collapsible tool call showing the custom agent name and active tool; expandable to reveal all tool calls, the prompt sent, and the returned result. Not checked yet Subagent chats are accessible as read-only peer chats in the Agents window via the Conversations dropdown, the running-subagents indicator, or Open Subagent in the delegation chat. Not checked yet AI credit usage per subagent is visible by hovering over its section in the chat response, providing cost transparency for delegated work. Not checked yet Deprecates the infer frontmatter property in favor of the new user-invocable and disable-model-invocation properties. Not checked yet
13 details in Everything
└──▷ BREAKING ON UPGRADE
! The infer frontmatter property is deprecated; replace with user-invocable and disable-model-invocation in custom agent .agent.md files.
VS Code agents gain a local memory tool and Copilot Memory integration for persisting context across sessions and workspaces.
Adds the chat.tools.memory.enabled setting to turn the built-in agent memory tool on or off. Not checked yet Adds chat.copilotMemory.enabled (experimental, disabled by default) to back repository memory with GitHub-hosted Copilot Memory instead of local files. Not checked yet New memory tool stores agent notes in three scoped local paths: /memories/ (user), /memories/repo/ (repository, workspace-scoped), and /memories/session/ (session, cleared when chat ends). Not checked yet User memory auto-loads the first 200 lines into the agent's context at the start of every session. Not checked yet New Chat: Show Memory Files command opens a browsable list of all memory files across scopes. Not checked yet
+4 more
show less
New Chat: Clear All Memory Files command removes all memory files across all scopes. Not checked yet Copilot Memory Preview is a GitHub-hosted repository-scoped memory system shared across Copilot cloud agent, Copilot code review, and Copilot CLI; memories auto-expire after 28 days. Not checked yet Repository owners can review and delete Copilot Memory entries under Repository Settings > Copilot > Memory on GitHub. Not checked yet The Plan agent persists its implementation plan to a plan.md file in session memory, viewable via Chat: Show Memory Files. Not checked yet
9 details in Everything
VS Code Agents window gains remote agent sessions over SSH and dev tunnels, plus a browser-based client at insiders.vscode.dev/agents
└──▷ HOW TO FIND IT
Connect to a remote host and start an agent session over SSH directly from the Agents window.
📍 In the Agents window, select New (⌘N / Ctrl+N), open the workspace dropdown, choose Remote › SSH, enter user@hostname, select a folder, type a prompt, and press Enter.
New Agents window lets you connect to remote machines and start or monitor agent sessions via SSH or dev tunnel, using the Agent Host Protocol (AHP AHP Analytic Hierarchy Process, a structured decision-making method that breaks a complex choice into a weighted hierarchy of criteria and scores alternatives against them, letting cyber tools rank or prioritize options mathematically. ); no extra agent installation required on the remote machine. Not checked yet Connect to a remote machine over SSH by selecting the Remote tab in the workspace dropdown, choosing SSH , and entering a user@hostname connection string — the VS Code CLI is installed and started automatically. Not checked yet Connect via dev tunnel by selecting Tunnels in the workspace dropdown and choosing an active tunnel, enabling remote agent sessions on machines running a dev tunnel. Not checked yet Browser-based Agents window available at https://insiders.vscode.dev/agents lets you manage agent sessions from any device without installing VS Code locally. Not checked yet Start a dev tunnel on a remote host with code-insiders tunnel (or code tunnel for the stable release) to expose it to the browser-based Agents window. Not checked yet
+1 more
show less
Hosts bar in the browser client shows connection status (Online /'Offline') for each tunnel host and automatically reconnects sessions when a host comes back online. Not checked yet
6 details in Everything
VS Code adds a preview artifacts panel surfacing screenshots, plans, and documents alongside chat, with three configurable extraction rules.
└──▷ USE IT
Enable the preview artifacts panel and add a rule to surface any HTML files the agent writes as a Web Pages artifact group.
json
copy
{
"chat.artifacts.enabled": true,
"chat.artifacts.rules.byFilePath": {
"**/*.html": { "groupName": "Web Pages" }
}
}
Disable all rules-based extraction so only agent-set and subagent artifacts appear in the panel.
json
copy
{
"chat.artifacts.rules.byMimeType": {},
"chat.artifacts.rules.byFilePath": {},
"chat.artifacts.rules.byMemoryFilePath": {}
}
New chat.artifacts.enabled setting (set to true) enables the preview artifacts panel, which surfaces screenshots, plans, and documents in a collapsible tree alongside the Chat view and Agents window. Not checked yet Adds chat.artifacts.rules.byMimeType setting to match tool results by MIME type pattern (e.g. 'image/*'), defaulting to { "image/*": { "groupName": "Screenshots", "onlyShowGroup": true } }. Not checked yet Adds chat.artifacts.rules.byFilePath setting to match workspace files by glob pattern (e.g. '**/*plan*.md'), defaulting to { "**/*plan*.md": { "groupName": "Plans" } }. Not checked yet Adds chat.artifacts.rules.byMemoryFilePath setting to match memory tool writes by glob pattern (e.g. '**/*plan*.md'), defaulting to { "**/*plan*.md": { "groupName": "Plans" } }. Not checked yet Artifacts panel groups outputs by origin: rules-based extraction, agent-set artifacts via the #artifacts tool, and subagent artifacts (each subagent shown in its own labeled group). Not checked yet
+1 more
show less
Agents can invoke the #artifactRules tool to override settings-based extraction rules for the duration of a session. Not checked yet
6 details in Everything
VS Code agents docs surface new config keys for checkpoint display, auto-accept delays, sensitive-file approval, and single-file diff toggling.
└──▷ USE IT
Protect workspace config and env files from silent agent edits by requiring manual approval only for those paths.
json
copy
// in settings.json
"chat.tools.edits.autoApprove": {
"**/*": true,
"**/.vscode/*.json": false,
"**/.env": false
}
Show a per-request file-change summary in chat so you can gauge impact before deciding whether to restore a checkpoint.
json
copy
// in settings.json
"chat.checkpoints.enabled": true,
"chat.checkpoints.showFileChanges": true
Open a focused single-file diff in the Agents window instead of the multi-file diff editor, useful when reviewing one changed file at a time.
json
copy
// in settings.json
"sessions.changes.openSingleFileDiff": true
Adds chat.checkpoints.showFileChanges setting (set to true) to display a changed-files summary with diff statistics after each completed agent request. Not checked yet Adds chat.checkpoints.enabled setting to control whether VS Code creates per-request file snapshots, enabling Restore Checkpoint and Fork Conversation workflows. Not checked yet Adds sessions.changes.openSingleFileDiff setting to open a focused single-file diff instead of the default multi-file diff editor when selecting a file in the Changes view. Not checked yet Adds chat.editRequests setting to configure or disable editing of previously sent chat prompts (which reverts and re-runs affected requests). Not checked yet Adds chat.tools.edits.autoApprove setting accepting glob patterns to require explicit approve/reject for sensitive files (e.g., **/.vscode/*.json, **/.env) before the agent applies edits. Not checked yet
+9 more
show less
Adds chat.editing.autoAcceptDelay setting to automatically accept pending extension-host edits after a configurable countdown delay. Not checked yet Adds chat.editing.revealNextChangeOnResolve setting (set to false) to stay in the current file instead of auto-opening the next file with pending edits. Not checked yet Introduces the Fork Conversation action on any checkpoint to spin off an independent agent session from that point in the conversation history. Not checked yet Introduces Mark as Reviewed per-file toolbar action in the Agents window Changes view to track reviewed files within a Branch Changes changeset. Not checked yet Introduces the Add Feedback / Submit Feedback inline commenting flow in the Changes view diff editor, letting practitioners annotate specific code ranges to direct agent follow-up edits. Not checked yet Introduces Redo action after restoring a checkpoint to recover changes that were undone. Not checked yet Adds Show Side by Side Diff , Show Inline Diff , Expand All Diffs , and Collapse All Diffs controls in the experimental single-pane Changes editor layout. Not checked yet Adds a sync button in the Files panel to pull upstream changes from the base branch before an agent session starts, reducing merge conflicts on integration. Not checked yet Adds session file-change statistics to the sessions list so practitioners can gauge an agent session's scope before opening it. Not checked yet
14 details in Everything
VS Code agents get granular approval controls: permission levels, per-tool/URL/terminal policies, sandboxing sandboxing Running code inside a restricted environment that limits what it can access or change. The isolation reduces the impact of mistakes or untrusted code, depending on the sandbox's controls. , and Autopilot mode.
└──▷ USE IT
Auto-approve outbound requests to your internal Contoso domains but still manually review fetched response content before it enters the chat context.
json
copy
// in settings.json
{
"chat.tools.urls.autoApprove": {
"https://*.contoso.com/*": {
"approveRequest": true,
"approveResponse": false
}
}
}
Adds chat.permissions.default setting to configure the default permission level (Default Approvals, Assisted permissions, or Bypass Approvals) for new agent chat sessions. Not checked yet Adds chat.assistedPermissions.enabled setting to expose the Assisted permissions level in the permissions picker, which uses an LLM LLM A large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. judge to auto-approve low-risk tool calls and prompt review for others. Not checked yet Adds chat.tools.eligibleForAutoApproval setting to control which tools are eligible for auto-approval; set a tool to false to force manual approval every time. Not checked yet Adds chat.tools.urls.autoApprove setting accepting a boolean or an object with approveRequest and approveResponse properties, supporting exact URLs, glob patterns, and wildcards to control pre- and post-approval of URL access. Not checked yet Adds chat.tools.terminal.autoApprove setting to extend or override the built-in terminal command allow/deny list, with support for exact command names (set to true/false) and regex patterns wrapped in / characters. Not checked yet
+7 more
show less
Adds chat.agent.sandbox.enabled setting to restrict file system and network access for agent commands. Not checked yet Adds chat.autopilot.advanced.enabled setting Preview to delegate task-completion decisions to a separate fast model instead of the agent itself, providing guided iteration in Autopilot mode. Not checked yet Adds Chat: Manage Tool Approval Command Palette command to centrally review and configure per-tool pre-approval (skip confirmation before run) and post-approval (skip reviewing tool output) grouped by source (MCP MCP Model Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. server or extension). Not checked yet Adds Chat: Reset Tool Confirmations Command Palette command to clear all saved tool approvals at once. Not checked yet Introduces Autopilot mode (selected from the agent mode picker) with continuous iteration, auto-approval of all tools, auto-retry on errors, and auto-response to blocking clarifying questions. Not checked yet Introduces a two-step URL approval flow for tools like #web/fetch: separate confirmation dialogs for the outbound request and the fetched response content, to mitigate prompt injection prompt injection An attack where text the model reads, such as a web page, file or tool result, carries instructions that override the user's. It is the main security risk for agents with tools. from user-generated content. Not checked yet Supports organization-level managed permission policies for shell commands, file operations, and domains that take precedence over session-level permission settings, including Bypass Approvals and Autopilot. Not checked yet
12 details in Everything
VS Code adds built-in browser tools for agents, enabling autonomous web app testing and interactive feedback loops without an external MCP MCP Model Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. server.
└──▷ HOW TO FIND IT
Let an agent run your dev server, exercise a user flow in the browser, and fix any failures it finds — all in one prompt.
📍 Start the development server and open the app in the browser.
Add a task named Review browser tools , mark it complete, and reload the page.
Verify that the completed task persists. If any step fails, fix the issue and repeat the flow.
Enable or verify the browser tools setting so agents can use the built-in browser capabilities.
json
copy
// settings.json
{
"workbench.browser.enableChatTools": true
}
Point the agent at a specific visual problem by attaching element-level comments from the rendered page instead of describing it in text.
📍 1. Open your application in the integrated browser.
2. Open the Add to Chat dropdown and select Comment on Elements .
3. Select the navigation menu and enter: Collapse this navigation into a menu button on narrow screens.
4. Select the form and enter: Align these fields and show validation errors below each input.
5. Send the prompt and ask the agent to implement the feedback and verify the result at desktop and mobile viewport sizes.
Adds workbench.browser.enableChatTools setting (on by default) to enable built-in browser tools for agents — no external MCP server required. Not checked yet Introduces built-in agent browser tool functions: openBrowserPage, navigatePage, readPage, screenshotPage, clickElement, hoverElement, dragElement, typeInPage, handleDialog, and runPlaywrightCode for custom Playwright automation. Not checked yet Adds Share with Agent action in the integrated browser toolbar, letting agents access an authenticated or stateful tab's cookies and storage. Not checked yet Adds Add to Chat dropdown in the browser toolbar with actions to attach page elements (with HTML, CSS, and optional screenshot), element-level comments, viewport/area/full-page screenshots, and console logs directly to agent prompts. Not checked yet Adds Comment on Elements action under Add to Chat to associate specific change requests with individual rendered elements on the page. Not checked yet
+4 more
show less
Adds BrowserChatTools enterprise policy so administrators can disable browser tools org-wide. Not checked yet Adds agent network filtering to restrict the domains that agent browser tools can reach. Not checked yet Agent-opened browser tabs use isolated ephemeral sessions (no shared cookies or storage); in the Agents window, tabs are isolated per session so an agent can only access its own tabs. Not checked yet In autopilot mode, VS Code automatically declines agent requests to share user-opened tabs, protecting session privacy. Not checked yet
9 details in Everything
VS Code agents gain browser tools, background terminal commands, and new settings for tool call grouping and terminal control
└──▷ USE IT
Keep tool calls visually separate from thinking output in long agent sessions to make it easier to audit each tool invocation individually.
json
copy
// In VS Code settings (settings.json)
{
"chat.agent.thinking.collapsedTools": "off"
}
Force terminal output to appear inline in chat so you can review command results without switching to the terminal panel.
json
copy
// In VS Code settings (settings.json)
{
"chat.tools.terminal.outputLocation": "inline"
}
Override the agent terminal shell on Windows to PowerShell for reliable shell integration instead of cmd.
json
copy
// In VS Code settings (settings.json)
{
"chat.tools.terminal.terminalProfile.windows": "PowerShell"
}
Adds chat.agent.thinking.collapsedTools setting Experimental to control tool call grouping in chat: off keeps tool calls separate, withThinking groups with thinking when available, and always always groups. Not checked yet Adds chat.agent.thinkingStyle setting: when set to collapsed, reasoning and grouped tool calls appear in separate collapsible sections. Not checked yet Adds chat.tools.terminal.outputLocation setting Experimental to configure whether terminal command output appears inline in chat or directly in the integrated terminal. Not checked yet Adds chat.tools.terminal.enforceTimeoutFromModel setting to control whether the agent-specified timeout for terminal commands is enforced. Not checked yet Adds chat.tools.terminal.terminalProfile.windows, chat.tools.terminal.terminalProfile.osx, and chat.tools.terminal.terminalProfile.linux settings to override the shell used by the agent terminal tool per platform. Not checked yet
+8 more
show less
Adds github.copilot.chat.virtualTools.threshold setting to automatically manage large tool sets and avoid the 128-tool-per-request limit. Not checked yet New built-in browser tools let agents open an integrated browser, exercise user flows, inspect page content and console errors, and fix discovered issues — without requiring an external MCP MCP Model Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. server. Not checked yet Agents can now push long-running terminal commands (e.g. dev servers, watch-mode builds) to the background via a Continue in Background button, allowing the agent to continue other tasks while the command runs. Not checked yet Background terminals that have not been revealed are automatically cleaned up when their command finishes, preventing stale terminal accumulation. Not checked yet Agents can now run commands directly in the background without user interaction; selecting the Show link in the chat tool invocation header reveals the terminal and keeps it open after completion. Not checked yet Tool input parameters are now editable before a tool runs: expand the tool confirmation dialog chevron, edit parameters, then select Allow to run with modified inputs. Not checked yet Tools can be explicitly referenced in prompts by typing # followed by the tool name (e.g. #web, #problems, #codebase); typing # alone shows all available tools including built-in, MCP, extension tools, and tool sets. Not checked yet Chat requests are capped at a maximum of 128 tools enabled at a time; exceeding this limit surfaces an actionable error with instructions to deselect tools or enable virtual tools. Not checked yet
13 details in Everything
VS Code agent harnesses agent harnesses The software around an AI agent that manages its tools, execution loop, state and limits. It determines how the model's proposed actions are run and checked. doc introduces Local, Copilot, Claude, and Codex session targets with isolation modes, remote control, and slash commands.
└──▷ USE IT
Copy a .env file and node_modules into every new worktree so Claude or Codex sessions have the credentials and dependencies they need.
json
copy
// In VS Code settings.json
{
"git.worktreeIncludeFiles": [".env", "node_modules"]
}
Monitor and steer a long-running Copilot coding session from your phone while the agent works in the background.
Use Claude without a GitHub sign-in by supplying an Anthropic API key so billing goes through your Anthropic account.
json
copy
// ~/.claude/settings.json
{
"env": {
"ANTHROPIC_API_KEY": "<your-key>"
}
}
Adds the Session Target control in the Chat view and Agents window for selecting among Local, Copilot, Claude, Codex, and Cloud agent harnesses when starting a new session. Not checked yet Introduces git.worktreeIncludeFiles setting to specify Git-ignored files and folders (e.g. .env files, installed dependencies) that VS Code copies into new worktrees. Not checked yet Adds chat.assistedPermissions.enabled setting to enable Assisted permissions mode for Copilot folder sessions alongside Default Approvals and Bypass Approvals. Not checked yet Enables remote control of Copilot sessions via /remote on slash command, synchronizing session history, tool activity, status, approvals, and questions with GitHub.com and the GitHub Mobile app. Not checked yet Adds github.copilot.chat.cli.remote.enabled setting to disable remote control support for Copilot sessions. Not checked yet
+12 more
show less
Adds github.copilot.chat.claudeAgent.enabled setting to enable or disable Claude agent harness agent harness The software around an AI agent that manages its tools, execution loop, state and limits. It determines how the model's proposed actions are run and checked. support in VS Code. Not checked yet Supports Claude authentication via ANTHROPIC_API_KEY in the environment or in the env object in ~/.claude/settings.json, and via CLAUDE_CODE_OAUTH_TOKEN with a token from claude setup-token. Not checked yet Adds chat.agentHost.allowSignedOutWhenUsable setting to open the Agents window while signed out of GitHub, showing only Anthropic-native models until sign-in. Not checked yet Adds github.copilot.chat.claudeAgent.allowDangerouslySkipPermissions setting to bypass all Claude permission checks (intended only for isolated sandbox environments without internet access). Not checked yet Adds chat.agentHost.codexAgent.enabled and chat.editor.codex.preferAgentHost settings to run Codex sessions experimentally on the Agent Host. Not checked yet Introduces Claude-specific slash commands including /agents, /hooks, /memory, /init, /pr-comments, /review, and /security-review for session and PR workflows. Not checked yet Introduces Copilot-specific slash commands including /compact (reduce conversation context), /research (start deep research), /yolo, and /autoApprove (control automatic tool approval). Not checked yet Adds a New Worktree isolation mode in the Agents window to run Copilot, Claude, and Codex sessions in an isolated Git worktree on a new branch, separate from the active workspace. Not checked yet Adds the /research command for Copilot sessions (preview, VS Code Insiders only) to run a read-only research agent that produces a Markdown report with citations from the codebase, GitHub repositories, and the web. Not checked yet Adds Chat: New Custom Agent Command Palette command to create workspace custom agents usable in Copilot sessions. Not checked yet Introduces Cloud session target for running agent sessions on the provider's remote infrastructure against a GitHub repository and pull request. Not checked yet Introduces Built-in Agent and Plan roles for Local harness sessions, switchable from the agent picker during a session. Not checked yet
17 details in Everything
VS Code gains a dedicated Agents window Preview for orchestrating parallel agent sessions across all workspaces from one place.
└──▷ TRY IT
Open the Agents window directly from the terminal to start orchestrating agent sessions without touching the editor UI.
Allow opening the Agents window without a GitHub sign-in when Claude is already configured with Anthropic credentials.
json
copy
// .vscode/settings.json or user settings
{
"chat.agentHost.allowSignedOutWhenUsable": true
}
Hide the empty Chats default group from the sessions list to reduce visual noise when no quick chats are active.
json
copy
// .vscode/settings.json or user settings
{
"sessions.list.showEmptyDefaultGroups": false
}
Adds the Agents window, a dedicated VS Code window for chat-first, agent-first workflows that manages sessions across all workspaces in parallel — accessible via Chat: Open Agents Window from the Command Palette, the Open in Agents title-bar button, or code --agents from the command line. Not checked yet Adds https://insiders.vscode.dev/agents as a browser-accessible entry point for Agents window sessions from any device. Not checked yet Adds chat.agentHost.allowSignedOutWhenUsable setting to enable opening the Agents window without GitHub sign-in when Claude is configured with Anthropic credentials (off by default). Not checked yet Adds sessions.list.showEmptyDefaultGroups setting (default true) to control whether empty default session groups such as Chats remain visible in the sessions list. Not checked yet Supports starting a background session without leaving the current one by pressing Alt+Enter (or holding Alt and selecting Send ) in the Agents window. Not checked yet
+6 more
show less
Adds quick chats — workspace-unscoped lightweight chats started via New Quick Chat from the Command Palette or ⌘K ⌘N / Ctrl+K Ctrl+N — for questions or tasks not tied to a project. Not checked yet Adds an integrated browser in the Agents window, openable via Open Integrated Browser from the Command Palette or by selecting a localhost link, with per-session tab and page state isolation. Not checked yet Adds a Tasks panel in the Agents window for configuring and running named tasks (e.g. npm run build, pytest) scoped to the active session's workspace or worktree, with an option to auto-run on worktree creation. Not checked yet Provides a Changes view with a dropdown to filter by branch changes, uncommitted changes, all changes, or changes from the last agent turn, plus range-based diff feedback for the agent. Not checked yet Supports choosing an agent harness agent harness The software around an AI agent that manages its tools, execution loop, state and limits. It determines how the model's proposed actions are run and checked. (Copilot, Claude, Codex) per session, with configurable model, permission level, and isolation mode. Not checked yet Sessions list supports grouping by workspace or time, custom groups, pinning, and drag-and-drop rearrangement. Not checked yet
11 details in Everything
VS Code adds agent sandboxing sandboxing Running code inside a restricted environment that limits what it can access or change. The isolation reduces the impact of mistakes or untrusted code, depending on the sandbox's controls. Preview with OS-level file system and network isolation for terminal commands, plus new network domain filtering settings.
└──▷ USE IT
Enable agent sandboxing on Linux so that terminal commands run during agent sessions cannot read from $HOME or make outbound network calls outside permitted domains.
json
copy
// In VS Code settings (settings.json)
{
"chat.agent.sandbox.enabled": "on",
"chat.agent.sandbox.allowNetwork": false,
"chat.agent.networkFilter": true,
"chat.agent.allowedNetworkDomains": ["registry.npmjs.org", "api.github.com"],
"chat.agent.deniedNetworkDomains": []
}
Install the required Linux packages (bubblewrap and socat) before enabling agent sandboxing on Debian/Ubuntu.
copy
$ sudo apt-get install bubblewrap socat
Adds agent sandboxing Preview that uses OS-level isolation to restrict file system and network access for terminal commands run during agent sessions, enforced via Apple Seatbelt on macOS and bubblewrap/socat on Linux/WSL2 WSL2 A Microsoft-maintained compatibility layer in Windows 10/11 that runs a real Linux kernel inside a lightweight VM, letting cyber tools built for Linux execute natively on Windows without dual-booting or a full VM. . Not checked yet Adds chat.agent.sandbox.enabled setting (accepts off (default) or on) to enable or disable agent sandboxing on macOS and Linux. Not checked yet Adds chat.agent.sandbox.allowNetwork setting to control whether sandboxed terminal commands have unrestricted outbound network access; when off, all outbound connections are blocked unless specific domains are explicitly allowed. Not checked yet Adds chat.agent.networkFilter setting to activate network domain filtering for agent tools (fetch tool, integrated browser) and sandboxed terminal commands, independently of sandboxing. Not checked yet Adds chat.agent.allowedNetworkDomains and chat.agent.deniedNetworkDomains settings to control which domains the agent and sandboxed commands can reach; deny rules always take precedence. Not checked yet
+6 more
show less
Introduces per-command read path grants: VS Code parses commands before execution and automatically grants read access to the specific paths needed, covering common workflows including git, node, npm, dotnet, Java, and Rust. Not checked yet Sandboxed terminal commands are auto-approved without a confirmation prompt because they run in a controlled OS-enforced environment, reducing approval fatigue during long agent sessions. Not checked yet Adds network domain trust boundaries: VS Code prompts for trust before the agent fetches content from a URL, integrated with the existing Trusted Domains list. Not checked yet Adds MCP MCP Model Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. server trust boundaries: VS Code prompts to trust each MCP server before it runs and re-prompts after configuration changes. Not checked yet Supports retry-with-network-access flow: when a sandboxed command is blocked by network restrictions, the agent first asks to retry inside the sandbox with unrestricted network access before falling back to running outside the sandbox. Not checked yet File system isolation defaults: read access allowed for workspace folders and sandbox runtime temp folder; reads from $HOME denied by default to protect SSH keys, shell config, and credentials; write access limited to current working directory and subdirectories. Not checked yet
11 details in Everything
VS Code introduces a dedicated Agent Host process with AHP AHP Analytic Hierarchy Process, a structured decision-making method that breaks a complex choice into a weighted hierarchy of criteria and scores alternatives against them, letting cyber tools rank or prioritize options mathematically. protocol, remote execution, shared sessions, and new code agent host CLI command.
└──▷ TRY IT
Start a remote standalone Agent Host exposed via dev tunnel so clients can connect from another machine.
copy
$ code agent host --tunnel
Enable the Agent Host process in VS Code settings so agent sessions run independently of the extension host.
json
copy
{
"chat.agentHost.enabled": true
}
Adds the chat.agentHost.enabled setting to enable the new dedicated Agent Host process, which runs agent sessions independently of editor windows and extension lifecycle. Not checked yet Introduces the Agent Host Protocol (AHP), an open JSON-RPC RPC Remote Procedure Call, a protocol that lets a program execute a function on another machine as if it were local, giving cyber tools a structured way to communicate across processes or hosts without managing raw sockets. protocol over WebSocket WebSocket A protocol that upgrades an HTTP connection into a persistent, full-duplex channel, letting a server push data to a client without repeated polling. Cyber tools use it for real-time event streaming and live command output. for communication between the Agent Host and clients, with immutable state and pure reducers for synchronized session data. Not checked yet Adds code agent host CLI command to start a standalone Agent Host server; supports --tunnel option to expose it through a dev tunnel. Not checked yet Agent Host reads harness-agnostic MCP MCP Model Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. config from .mcp.json (workspace) and ~/.copilot/mcp-config.json (user) instead of .vscode/mcp.json directly. Not checked yet Agent Host reads user-level agent customizations from ~/.copilot and ~/.claude folders rather than VS Code profile user data. Not checked yet
+5 more
show less
Enables shared multi-window sessions, multiple chats per session, quick chats, and remote hosting — capabilities available only on the Agent Host. Not checked yet Introduces Assisted permissions level, available only on the Agent Host. Not checked yet On the Agent Host, Autopilot is an agent mode (not a permission level as in the extension host), and edits are applied directly to the session folder or worktree for diff-based review. Not checked yet Agent sessions persist independently of editor windows — closing a window and reopening it from another window resumes the session while the Agent Host remains running. Not checked yet Supports remote execution: the Agent Host can run as a standalone server on a remote machine, reachable via SSH or dev tunnel, keeping workspace operations close to source code. Not checked yet
10 details in Everything
└──▷ BREAKING ON UPGRADE
! Agent Host sessions apply edits directly to the session folder or worktree (reviewed as diffs before commit/merge/discard), replacing the extension-host behavior of tracking edits as pending until kept or undone.! The Agent Host does not read .vscode/mcp.json directly; MCP servers requiring interactive input (e.g. ${input:...} variables) are not forwarded to the Agent Host.! Customizations stored only in VS Code profile user data are a legacy location that the Copilot agent on the Agent Host does not read — user-level customizations must be placed in ~/.copilot or ~/.claude.
VS Code agents gain a browser-integrated feedback loop so agents can run, exercise, inspect, and verify web apps autonomously.
Adds browser tools integration for agents, enabling an agent to run a web app, exercise user flows in an integrated browser, inspect results, fix problems, and verify changes — forming a visual and interactive feedback loop. Not checked yet
1 detail in Everything
advisory CVE-2026-69320
Aug 11, 2026
·
issue 128
CVE-2026-69320: Visual Studio Code web deployments - environmentService.ts - RCE RCE Remote code execution, the ability to run attacker-chosen code on another system through a vulnerability. Its impact depends on the permissions and access that code receives. via NODE_OPTIONS --import from URL payload (CVSS CVSS Common Vulnerability Scoring System, a method for describing a vulnerability's characteristics and severity. Its score is not the probability that attackers will exploit it. 8.8, affects < 1.132.1)︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄ Not checked yet
VS Code enterprise policies add Claude agent toggle and four new chat restriction controls for managed environments.
Adds sessions.chat.claudeAgent.enabled policy to control access to the Claude agent in managed environments. Not checked yet Adds ChatStrictPluginOnlyCustomization policy that blocks standalone user and workspace skills, agents, hooks, instructions, and MCP MCP Model Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. servers while keeping eligible plugin customizations available. Not checked yet Adds ChatAllowManagedHooksOnly policy that restricts hooks to enterprise-managed sources and policy-force-enabled plugins only. Not checked yet Adds ChatAllowManagedMcpServersOnly policy that limits MCP server execution to the enterprise-managed MCP allowlist. Not checked yet
4 details in Everything
New workbench.browser.newTabPlacement setting controls where integrated browser tabs open in code-oss.
VS Code 1.131 adds subagent tracking at a glance, voice dictation across the editor, and agent-actionable Markdown comments.
Track running subagents at a glance with a new subagent status view. Not checked yet Dictate text across VS Code using voice input. Not checked yet Edit Markdown with agent-actionable comments that agents can act on directly. Not checked yet
3 details in Everything
VS Code enterprise policies gain MCP MCP Model Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. server allowlist/denylist controls and Copilot Chat OpenTelemetry OpenTelemetry An open standard and set of tools for collecting traces, metrics and logs from applications. It lets services export observability data without tying instrumentation to one monitoring vendor. export settings.
Adds ChatAllowedMcpServers policy to allowlist which Model Context Protocol Model Context Protocol Model Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. servers can be installed or run. Not checked yet Adds ChatDeniedMcpServers policy to denylist MCP servers that must be blocked, with deny rules taking precedence over the allow list. Not checked yet Adds OpenTelemetry export configuration policies for Copilot Chat, including github.copilot.chat.otel.enabled, exporterType, otlpEndpoint, protocol, serviceName, resourceAttributes, headers, captureContent, and outfile. Not checked yet
3 details in Everything
VS Code enterprise policies gain MCP MCP Model Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. server allowlist/denylist controls and Copilot Chat OpenTelemetry OpenTelemetry An open standard and set of tools for collecting traces, metrics and logs from applications. It lets services export observability data without tying instrumentation to one monitoring vendor. export settings.
Adds ChatAllowedMcpServers policy to allowlist which Model Context Protocol Model Context Protocol Model Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. servers can be installed or run. Not checked yet Adds ChatDeniedMcpServers policy to denylist MCP servers that must be blocked, with deny rules taking precedence over the allow list. Not checked yet Adds OpenTelemetry export configuration policies for Copilot Chat, including github.copilot.chat.otel.enabled, exporterType, otlpEndpoint, protocol, serviceName, resourceAttributes, headers, captureContent, and outfile. Not checked yet
3 details in Everything
VS Code 1.130: agent sessions in dedicated processes, assisted tool approvals, multi-file diff review, and Git diff file opening.
Runs agent sessions in a dedicated process for improved isolation and stability. Not checked yet Reduces approval interruptions with assisted tool approvals in agent workflows. Not checked yet Enables faster multi-file change review directly in the Agents window. Not checked yet Supports opening files directly from Git diffs in the terminal. Not checked yet
4 details in Everything
VS Code 1.130: agent sessions in dedicated processes, assisted tool approvals, multi-file diff review, and Git diff file opening.
Runs agent sessions in a dedicated process for improved isolation and stability. Not checked yet Reduces approval interruptions with assisted tool approvals in agent workflows. Not checked yet Enables faster multi-file change review directly in the Agents window. Not checked yet Supports opening files directly from Git diffs in the terminal. Not checked yet
4 details in Everything
Adds Auto mode switch in weekly Copilot limit messaging
└──▷ GET THIS VERSION
copy
$ git clone --branch v0.45.1 https://github.com/microsoft/vscode.git
$ git checkout v0.45.1
VS Code 1.111 adds agent permissions, agent-scoped hooks, and improved agent debugging to the Copilot agent experience.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.111.0 https://github.com/microsoft/vscode.git
$ git checkout 1.111.0
Introduces agent permissions, giving users control over what actions AI agents are allowed to perform. Not checked yet Adds agent-scoped hooks, enabling lifecycle callbacks tied to individual agent sessions. Not checked yet Improves agent debugging and troubleshooting tooling for diagnosing agent behavior. Not checked yet
3 details in Everything
VS Code 1.110 adds agent plugins, agentic browser control, session memory, context compaction context compaction Condensing an AI conversation or working history to free space in the model's context window. A summary preserves useful information, but details can be lost. , and session forking for AI workflows.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.110 https://github.com/microsoft/vscode.git
$ git checkout 1.110
Adds Agent Plugins Experimental : install prepackaged bundles of skills, tools, and hooks from the Extensions view to extend agent capabilities. Not checked yet Adds Agentic Browser Tools Experimental : lets the agent drive the integrated browser to interact with your app and verify its own changes autonomously. Not checked yet Adds Session Memory for plans: persists plans and guidance across conversation turns so agents maintain context in longer-running tasks. Not checked yet Adds Context Compaction: manually compact conversation history to free up context window context window The maximum amount of text, counted in tokens, a model can take in and produce in one request. Everything the model knows about the task has to fit inside it. space during long sessions. Not checked yet Adds Fork a Chat Session: creates a new, independent session that inherits conversation history, enabling exploration of alternative approaches without losing prior context. Not checked yet
+4 more
show less
Adds Agent Debug Panel Preview : provides real-time visibility into agent events, tool calls, and loaded customizations. Not checked yet Adds Create Agent Customizations from Chat: generates prompts, skills, agents, and hooks directly from a conversation. Not checked yet Adds Kitty graphics protocol Kitty graphics protocol A terminal graphics protocol created by the Kitty terminal emulator project that lets applications render pixel-accurate images directly inside a terminal, giving cyber tools rich visual output without a GUI. support in the integrated terminal: renders high-fidelity images inline in the terminal. Not checked yet Improves chat accessibility with screen reader enhancements, keyboard navigation, and notification signals. Not checked yet
9 details in Everything
VS Code 1.109.0 adds multi-agent orchestration, Claude/Anthropic support, MCP MCP Model Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. Apps, Copilot Memory, terminal sandboxing sandboxing Running code inside a restricted environment that limits what it can access or change. The isolation reduces the impact of mistakes or untrusted code, depending on the sandbox's controls. , and an integrated browser.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.109.0 https://github.com/microsoft/vscode.git
$ git checkout 1.109.0
Adds Claude agent support and new Anthropic model capabilities in chat. Not checked yet Introduces MCP Apps for rich, structured chat interactions. Not checked yet Adds Copilot Memory so agents retain context across sessions. Not checked yet Supports external indexing for faster code search in agent workflows. Not checked yet Introduces Agent Skills and organization-wide agent customizations for consistent results. Not checked yet
+5 more
show less
Enables agent orchestration so users can build custom multi-agent workflows. Not checked yet Adds background and cloud agent session management with the ability to delegate and resume tasks. Not checked yet Adds terminal command sandboxing and auto-approval rules for safer agent-driven terminal use. Not checked yet Introduces an integrated browser for testing web apps without leaving the editor. Not checked yet Revamps editor inline chat with faster streaming and improved reasoning result display. Not checked yet
10 details in Everything
VS Code 1.107 introduces Agent HQ with background agents and org-level custom agent sharing via GitHub Copilot.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.107.0 https://github.com/microsoft/vscode.git
$ git checkout 1.107.0
New Agent HQ hub lets you run multiple coding agents (background, cloud, or local) simultaneously alongside GitHub Copilot. Not checked yet Background agents run tasks in isolated workspaces managed automatically by VS Code. Not checked yet Organizations can now define and share custom agents across the team at the org level. Not checked yet
3 details in Everything
VS Code 1.106 adds plan agent for complex tasks, custom MCP MCP Model Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. registry for orgs, and Linux JSON policy management.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.106.0 https://github.com/microsoft/vscode.git
$ git checkout 1.106.0
New plan agent enables planning and coordinating complex multi-step coding tasks, with a session monitor for background agents. Not checked yet Inline suggestions engine is now open source. Not checked yet Adds custom MCP registry support so organizations can manage and restrict MCP server access. Not checked yet Adds Linux policy support — manage VS Code device policies via JSON policy files. Not checked yet Text selection now works on deleted code lines in the diff editor. Not checked yet
+1 more
show less
Refreshed, modernized product icon set across the UI. Not checked yet
6 details in Everything
VS Code 1.105 adds MCP MCP Model Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. marketplace, AI merge conflict resolution, OS notifications, and macOS native auth.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.105.0 https://github.com/microsoft/vscode.git
$ git checkout 1.105.0
New MCP marketplace preview lets users browse and install MCP servers directly from within the editor. Not checked yet Adds AI-assisted merge conflict resolution to help automatically resolve conflicts in the editor. Not checked yet Adds OS-level notifications for task completion and chat responses so you can context-switch freely. Not checked yet Supports macOS native broker (system authentication dialog) for Microsoft account sign-in. Not checked yet New recent chats panel lets users resume previous chat sessions from where they left off Experimental . Not checked yet
+1 more
show less
Supports fully-qualified tool names to avoid conflicts when multiple MCP servers expose identically named tools. Not checked yet
6 details in Everything
VS Code 1.104 adds auto model selection, sensitive-file edit confirmation, terminal auto-approve for agents, and AGENTS.md context support.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.104.0 https://github.com/microsoft/vscode.git
$ git checkout 1.104.0
Adds auto-model selection Preview : VS Code automatically picks the best available AI model for each request. Not checked yet New Language Model Chat Provider API lets extensions contribute their own AI models directly into VS Code. Not checked yet New confirm-edits-to-sensitive-files prompt requires user approval before AI agents modify security-sensitive files. Not checked yet New terminal auto-approve setting lets AI agents run terminal commands within defined safe rules without per-command prompts. Not checked yet Improved changed-files experience in chat reduces visual noise when reviewing AI-driven file edits. Not checked yet
+1 more
show less
Experimental support for AGENTS.md files automatically injects project-level context into chat sessions. Not checked yet
6 details in Everything
code-oss 1.103.1 adds GPT-5 prompt improvements and GPT-5 mini model support in Copilot Chat.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.103.1 https://github.com/microsoft/vscode.git
$ git checkout 1.103.1
Adds GPT-5 prompt improvements for enhanced Copilot Chat interactions. Not checked yet Supports GPT-5 mini as a selectable model in Copilot Chat. Not checked yet
2 details in Everything
VS Code 1.103 adds GPT-5, chat checkpoints, Git worktree support, and tool grouping for MCP MCP Model Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. agents.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.103.0 https://github.com/microsoft/vscode.git
$ git checkout 1.103.0
Adds GPT-5 model availability for use in VS Code chat. Not checked yet Adds chat checkpoints to restore a conversation to a previous good state. Not checked yet Adds Git worktree support to check out multiple branches simultaneously. Not checked yet Revamps the MCP tool picker experience with improved tool selection UI. Not checked yet Enables more than 128 tools per agent request via experimental tool grouping. Not checked yet
+1 more
show less
Adds a dedicated Chat Sessions view Experimental for managing coding agent sessions. Not checked yet
6 details in Everything
VS Code 1.102 ships GA MCP MCP Model Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. support, Copilot coding agent delegation, custom chat modes, and terminal auto-approval.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.102.0 https://github.com/microsoft/vscode.git
$ git checkout 1.102.0
MCP (Model Context Protocol Model Context Protocol Model Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. ) support is now generally available, with a dedicated MCP view and gallery for installing and managing MCP servers. Not checked yet MCP servers are now first-class resources in profiles and Settings Sync. Not checked yet Delegates background tasks to a Copilot coding agent session Preview . Not checked yet Generates custom instructions that reflect a project's conventions. Not checked yet Adds custom chat modes for tailoring chat to tasks like planning or research. Not checked yet
+3 more
show less
Enables automatic approval of selected terminal commands Experimental . Not checked yet Allows editing and resubmitting previous chat requests Experimental . Not checked yet Adds scroll-on-middle-click behavior in the editor. Not checked yet
8 details in Everything
VS Code 1.101 adds MCP MCP Model Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. auth, prompts/resources/sampling, debug mode, tool sets, and GitHub Copilot Coding Agent integration.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.101.0 https://github.com/microsoft/vscode.git
$ git checkout 1.101.0
Adds MCP support for prompts, resources, and sampling to expand agent coding workflows. Not checked yet Enables authentication for MCP servers that require it. Not checked yet New MCP development mode for debugging MCP servers. Not checked yet Supports publishing MCP servers directly from an extension via new extension APIs. Not checked yet Introduces Chat tool sets to group and manage related tools together. Not checked yet
+2 more
show less
New Source Control History Item Details view for browsing files changed in a commit. Not checked yet Integrates GitHub Copilot Coding Agent assignment and tracking directly within VS Code. Not checked yet
7 details in Everything
VS Code 1.100 adds custom chat instructions, GitHub/extension/notebook tools, MCP MCP Model Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. Streamable HTTP Streamable HTTP An MCP transport protocol that lets a client send JSON-RPC requests over HTTP POST while the server streams responses back using Server-Sent Events, replacing the older stdio and SSE transports for remote tool communication. , and staged-change diff decorations.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.100.0 https://github.com/microsoft/vscode.git
$ git checkout 1.100.0
Adds custom instructions and reusable prompt files for Chat, enabling repeatable, shareable AI workflows. Not checked yet Adds built-in Chat tools for searching GitHub repository code, querying extensions, and working with notebooks. Not checked yet Adds MCP support for Streamable HTTP transport and image content in Model Context Protocol Model Context Protocol Model Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. sessions. Not checked yet Introduces conversation summary and prompt caching to deliver faster responses on repeat Chat requests. Not checked yet Speeds up agent mode edits for quicker AI-driven code changes. Not checked yet
+2 more
show less
Improves multi-window support with floating window modes for Chat and editors. Not checked yet Adds Quick Diff decorations for staged changes, making staged hunks visually distinct in the editor gutter. Not checked yet
7 details in Everything
VS Code 1.99 adds stable agent mode with MCP MCP Model Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. server support, BYOK BYOK Bring Your Own Key — a cloud deployment model where the customer supplies and controls the encryption keys rather than the provider, so sensitive data stays under the customer's cryptographic authority. model access, and Next Edit Suggestions GA.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.99.0 https://github.com/microsoft/vscode.git
$ git checkout 1.99.0
Enables agent mode in VS Code Stable via chat.agent.enabled:true setting. Not checked yet Adds Model Context Protocol Model Context Protocol Model Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. (MCP) server support to extend agent mode with external tools. Not checked yet Introduces built-in agent mode tools for fetching web content, finding symbol references, and deep thinking. Not checked yet Next Edit Suggestions (AI-powered next-edit predictions) is now generally available. Not checked yet Adds Bring Your Own Key (BYOK) preview, letting users supply their own API keys to access additional language models in chat. Not checked yet
+3 more
show less
Introduces a unified chat experience with easy switching between ask, edit, and agent mode. Not checked yet Adds instant remote workspace indexing for faster and more accurate workspace searches in chat. Not checked yet Extends edit and agent mode to notebook editing, enabling AI-assisted creation and editing of notebooks. Not checked yet
8 details in Everything
VS Code 1.98 adds Copilot agent mode, terminal IntelliSense, notebook edits, diagnostics commit hook, and custom title bar on Linux.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.98.0 https://github.com/microsoft/vscode.git
$ git checkout 1.98.0
New Next Edit Suggestions Preview : Copilot predicts and surfaces the next edit you are likely to make in your code. Not checked yet New Agent mode Preview : Copilot autonomously completes multi-step tasks without manual prompting for each step. Not checked yet New Copilot Edits support for Jupyter notebooks, enabling iterative AI-assisted edits directly in notebook cells. Not checked yet New advanced codebase search in Copilot Chat lets Copilot find relevant files across the repo to ground chat responses. Not checked yet New Terminal IntelliSense Preview provides rich, context-aware completions inside the integrated terminal. Not checked yet
+5 more
show less
New drag-and-drop support for Peek References lets you open a reference directly in a new editor by dragging it. Not checked yet Custom title bar on Linux is now enabled by default, bringing consistent native-style chrome to Linux users. Not checked yet New Unresolved Diagnostics commit hook Preview prompts developers when committing with outstanding errors or warnings. Not checked yet Soft-delete for untracked files in source control: discarding untracked changes now moves files to the system trash instead of permanently deleting them. Not checked yet Custom Instructions for Copilot is now generally available, allowing persistent, per-repo or per-user behavior tailoring. Not checked yet
10 details in Everything
VS Code 1.97 adds Copilot next-edit predictions, extension publisher trust, compound logs, git blame, and Python no-config debug.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.97.0 https://github.com/microsoft/vscode.git
$ git checkout 1.97.0
Adds Copilot Next Edit Suggestions Preview : AI predicts and surfaces the next edit you are likely to make in the editor. Not checked yet Adds auto-accept for Copilot edits after a configurable timeout, reducing manual confirmation steps. Not checked yet Enables extension publisher trust, allowing practitioners to restrict which publishers' extensions are trusted in their environment. Not checked yet Adds compound logs view to combine multiple log sources into a single aggregated Output panel. Not checked yet Adds filtering for the Output panel, letting you search and narrow log contents inline. Not checked yet
+5 more
show less
Adds rich git blame information with the ability to open the blamed commit directly on GitHub. Not checked yet Adds filter and search on values in the Debug Variables panel to quickly locate specific runtime values. Not checked yet Adds inline values display for notebook code cell variables upon cell execution. Not checked yet Adds Python no-config debug mode to immediately debug a Python script or module without any launch configuration setup. Not checked yet Enables repositioning of the Command Palette and Quick Inputs by dragging them to a new position on screen. Not checked yet
10 details in Everything
VS Code 1.96 adds overtype mode, Copilot debug terminal command, extension allow lists, and attributable test coverage.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.96.0 https://github.com/microsoft/vscode.git
$ git checkout 1.96.0
Adds overtype mode, letting users toggle between overwrite and insert editing in the editor. Not checked yet Automatically adds missing TypeScript/JavaScript imports when pasting code. Not checked yet Introduces attributable test coverage, enabling filtering of code covered by a specific test. Not checked yet Enables moving views between the Primary and Secondary Side Bar. Not checked yet Adds ligature support in the integrated terminal. Not checked yet
+4 more
show less
Adds an extension allow list feature so organizations can configure which extensions are permitted to be installed. Not checked yet Introduces the copilot-debug terminal command to start a Copilot-assisted debugging session. Not checked yet Supports adding symbols and folders as context in Copilot Chat and Edits. Not checked yet Enables moving a Chat session directly into Copilot Edits to apply code suggestions. Not checked yet
9 details in Everything
VS Code 1.95 adds Copilot Edits for multi-file changes, code review, multiple GitHub accounts, and Pylance docstring generation.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.95.0 https://github.com/microsoft/vscode.git
$ git checkout 1.95.0
New Copilot Edits mode lets you iterate on large code changes spanning multiple files in a single session. Not checked yet Copilot Chat can now be pinned to the Secondary Side Bar so it stays visible while you work. Not checked yet Supports logging in to multiple GitHub accounts simultaneously within VS Code. Not checked yet New Copilot code review capability offers both a quick review pass and a deeper review of uncommitted changes. Not checked yet Pylance can now generate docstring templates for Python classes and methods. Not checked yet
+1 more
show less
Settings editor now shows an indicator distinguishing experimental and preview settings. Not checked yet
6 details in Everything
VS Code 1.94 adds Explorer search, Source Control Graph filters, Python test coverage, and Copilot in the Native REPL.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.94.0 https://github.com/microsoft/vscode.git
$ git checkout 1.94.0
Adds improved Find control in the Explorer view for quickly locating files. Not checked yet Adds more filtering options and interactivity to the Source Control Graph. Not checked yet Supports running Python tests with coverage and displays rich results inline in the editor. Not checked yet Faster startup via migration to ES Modules ES Modules The official JavaScript module system, standardized by TC39, that lets files import and export values using `import`/`export` syntax; tools adopt it for native browser compatibility and static analysis of dependencies. (ESM). Not checked yet Enables per-extension account preference, letting users specify which account an extension should use. Not checked yet
+3 more
show less
Adds Copilot inline chat and code completions inside the Python Native REPL. Not checked yet Supports drag-and-drop files and IntelliSense-driven context attachment in Copilot chat for more relevant responses. Not checked yet Adds automated test framework setup assistance to help configure a test environment for a workspace. Not checked yet
8 details in Everything
VS Code 1.93 adds a unified Profiles editor, Django test support, Copilot context improvements, and full IntelliSense on vscode.dev.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.93.0 https://github.com/microsoft/vscode.git
$ git checkout 1.93.0
New Profiles editor lets you switch and manage all profiles from a single consolidated view. Not checked yet Django unit test support enables discovering and running Django unit tests directly from the Test Explorer. Not checked yet Full project IntelliSense and package typings now available on vscode.dev for JavaScript and TypeScript. Not checked yet Notebook diff viewer can collapse unchanged cells to streamline change review. Not checked yet Table columns can now be resized via the keyboard for improved accessibility. Not checked yet
+3 more
show less
Source Control Graph view can be hidden, collapsed, or repositioned. Not checked yet GitHub Copilot gains context-adding in Quick Chat, improved test generation, and chat history. Not checked yet Experimental Copilot Inline Chat uses recent coding files as additional context. Not checked yet
8 details in Everything
VS Code 1.92 adds GPT-4o Copilot, panel-top positioning, CSS paste-to-reference, profile overrides, and configurable default browser.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.92.0 https://github.com/microsoft/vscode.git
$ git checkout 1.92.0
Enables configuring a preferred browser for opening links directly from VS Code. Not checked yet Adds one-click revert PR creation for merged GitHub Pull Requests. Not checked yet Adds granular auto-update controls for extensions via an improved update experience. Not checked yet Supports overriding an existing profile using the Profiles Editor. Not checked yet Supports pasting or drag-and-dropping files in CSS to automatically insert file references. Not checked yet
+2 more
show less
Adds option to position the Panel section at the top of the workbench. Not checked yet Upgrades GitHub Copilot Chat to OpenAI GPT-4o. Not checked yet
7 details in Everything
VS Code 1.91 introduces Profiles Editor preview, GitHub Copilot Chat API for extensions, and native REPL smart send.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.91.0 https://github.com/microsoft/vscode.git
$ git checkout 1.91.0
Adds Preview: Profiles Editor for managing profiles in a single place Not checked yet Brings Chat and Language Model APIs to Stable for extension development Not checked yet Enables Smart Send in the native REPL to run code chunks smoothly Not checked yet Introduces source control visualization of incoming and outgoing changes Not checked yet Adds enhanced Python environment discovery via python-environment-tools Not checked yet
+1 more
show less
Expands custom tab label variables and allows multiple extensions Not checked yet
6 details in Everything
VS Code 1.90 adds editor tabs multi-select, custom profiles, and Chat/Language Model API for extensions.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.90.0 https://github.com/microsoft/vscode.git
$ git checkout 1.90.0
└──▷ HOW TO FIND IT
Scan a specific notebook cell by restricting the find command to only that selection.
📍 In the notebook editor, select a cell and press Ctrl+F to enable Find in cell selection .
Adds profiles feature to open new windows with a preferred profile configuration Not checked yet Adds Chat and Language Model API for extension developers to build AI features into extensions Not checked yet Adds VS Code Speech integration to automatically read out Copilot Chat responses via text-to-speech Not checked yet Adds ability to restrict search to selected cells in notebooks via Find in cell selection Not checked yet Adds ability to attach different types of context to chat messages Not checked yet
+1 more
show less
Adds IntelliSense support within chat response code blocks for generated code Not checked yet
6 details in Everything
VSCode 1.89 adds hover-to-preview media in Markdown, Quick Search, and workspace-local extensions
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.89.0 https://github.com/microsoft/vscode.git
$ git checkout 1.89.0
Adds hoverToPreviewImagesAndVideosInMarkdown setting to enable image and video previews when hovering over links in Markdown files Not checked yet Adds restoreOpenEditorsOnBranchSwitch setting to automatically save and restore open editors when switching Git branches Not checked yet Enables middle-click paste support in the integrated terminal via the new terminal.integrated.enableMiddleClickPaste configuration key Not checked yet Introduces Quick Search feature allowing text search across the entire workspace without opening a file explorer or find panel Not checked yet Adds support for Local workspace extensions that are included and installed directly within the .vscode/extensions.json workspace configuration file Not checked yet
+1 more
show less
Adds contentExclusions configuration key to exclude specific files from Copilot's context analysis Not checked yet
6 details in Everything
VS Code 1.88 adds native Test Coverage API, terminal inline chat preview, and editor labels
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.88.0 https://github.com/microsoft/vscode.git
$ git checkout 1.88.0
Adds preview of Terminal inline chat via Copilot to start conversations directly from the terminal. Not checked yet Adds ability to apply custom editor labels to distinguish between editors with the same file names. Not checked yet
2 details in Everything
Code-oss 1.87.0 adds voice dictation, multi-cursor inline suggestions, Copilot rename/refactoring features, and sticky scroll.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.87.0 https://github.com/microsoft/vscode.git
$ git checkout 1.87.0
└──▷ HOW TO FIND IT
Activate the new voice dictation mode in the editor to input text hands-free.
📍 Open Command Palette (Ctrl+Shift+P), type Dictation , and select Toggle Voice Dictation .
Review and accept inline code suggestions for multiple cursors at once.
📍 Place multiple cursors in your code and type to trigger multi-cursor inline completions, then press Tab to accept.
Rename a symbol across files using Copilot's AI-powered suggestions.
📍 Right-click a symbol, select Refactor , then Rename and use the provided Copilot suggestion.
Preview refactoring changes across multiple files using the new side-by-side multi-diff editor.
📍 Initiate a refactor, then select Preview in Multi-Diff Editor to review changes before applying.
Adds Voice dictation in editor to type directly via speech input Not checked yet Enables Multi-cursor inline suggestions for reviewing and accepting completions across multiple cursors simultaneously Not checked yet Introduces Copilot-powered rename suggestions for symbol renaming Not checked yet Adds Side-by-side preview refactoring using the multi-diff editor to preview code changes before applying them Not checked yet Implements Sticky scroll in editor enabled by default to keep line numbers and folding icons visible while scrolling Not checked yet
5 details in Everything
VSCode 1.86 adds per-window zoom, triggered breakpoints, multi-file diffs, and voice commands
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.86.0 https://github.com/microsoft/vscode.git
$ git checkout 1.86.0
Adds window.zoomLevel support for per-window zoom levels Not checked yet Adds Triggered Breakpoints allowing breakpoint dependencies during debugging Not checked yet Enables Multi-file diff editor to review diffs across multiple files simultaneously Not checked yet Adds Hey Code voice command to start a chat session via voice Not checked yet Expands Sticky Scroll support to tree views and notebooks Not checked yet
+4 more
show less
Adds Markdown paste options for links, video, and audio elements Not checked yet Adds flexible Auto Save options to skip on errors or target specific file types Not checked yet Adds Source Control input customization and per-language editor settings Not checked yet Allows disabling notifications per extension via Extension Notifications Not checked yet
9 details in Everything
VS Code 1.85 adds floating editor windows, Python type hierarchy, heap snapshot viewing, and Sticky Scroll in trees.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.85.0 https://github.com/microsoft/vscode.git
$ git checkout 1.85.0
Python python.typeHierarchy.enabled setting enables a tree view for navigating class inheritance and type relationships. Not checked yet JavaScript debugger gains the ability to visualize heap snapshots, including memory object graphs, via the dedicated Heap Snapshot view. Not checked yet Source Control panel adds incoming/outgoing commit views (git.detectGitGraph) to review pending pull requests and changes directly in the IDE. Not checked yet Terminal and tree views support Sticky Scroll headers Preview for persistent context while scrolling large files or directories. Not checked yet Inline Chat now supports Rust code explanation alongside existing language support. Not checked yet
5 details in Everything
VS Code 1.84 adds audio cues, configurable activity bar position, editor tab controls, and Gradle support.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.84.0 https://github.com/microsoft/vscode.git
$ git checkout 1.84.0
Adds audio cues to indicate clear, save, and format activity via settings for accessibility. Not checked yet
1 detail in Everything
New UI density options, profile icons, Notebook symbol navigation, and Copilot test generation in VS Code 1.83.0
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.83.0 https://github.com/microsoft/vscode.git
$ git checkout 1.83.0
steps: In the Command Palette, type a command name to see a list of similar commands for easier discovery. Not checked yet config: Set workbench.profile.icon in your user or workspace settings JSON to display a custom icon for a profile. Not checked yet shell: Use the Notebooks: Go to Symbol command (default shortcut Ctrl+Shift+G on Windows/Linux, Cmd+Shift+G on Mac) within a notebook to jump to code symbols. Not checked yet config: Set python.debugging.stepSystem to false in your launch configuration or settings to configure whether the Python debugger steps into system/library code during debugging. Not checked yet Preview feature: GitHub Copilot now supports generating tests based on current framework and project conventions. Not checked yet
5 details in Everything
VS Code 1.82 adds built-in port forwarding, WebAssembly WebAssembly A binary instruction format that runs in browsers and server runtimes at near-native speed, letting developers compile C, Rust, or other languages into a portable, sandboxed module any compliant host can execute. debugging, sticky scroll updates, and more.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.82.0 https://github.com/microsoft/vscode.git
$ git checkout 1.82.0
Introduces webassembly as a supported debug target type for stepping through and setting breakpoints in WASM WASM A binary instruction format for a stack-based virtual machine, maintained by the W3C, that lets code compiled from any language run at near-native speed inside a sandboxed environment across browsers and servers. code Not checked yet Moves the Command Center to be visible by default in the title bar for quick file opening and command execution Not checked yet
2 details in Everything
VS Code 1.81 introduces accessible view for notifications/chats, git symlink support, and GitHub PR creation UX improvements
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.81.0 https://github.com/microsoft/vscode.git
$ git checkout 1.81.0
└──▷ HOW TO FIND IT
Use the new iterative /fix command in GitHub Copilot Chat to resolve issues iteratively.
📍 In the Copilot Chat panel, type /fix followed by a description of the error context to trigger automated fixes.
Adds Accessible View support for notifications, chat responses, and hovers to improve screen reader compatibility Not checked yet Enables Git repositories with symbolic links by supporting repository paths containing symlinks Not checked yet Introduces partial profiles in VS Code Profiles, allowing Apply to all profiles options for finer configuration control Not checked yet Adds collapse functionality for unchanged regions in the diff editor and improves diff region text alignment Not checked yet Allows searching text in closed notebooks and adds sticky scroll display for Markdown headers in notebook cells Not checked yet
+4 more
show less
Implements error-tolerant pytest discovery that continues across all files for Python test discovery Not checked yet Enables connecting to remote Jupyter servers within GitHub Codespaces for direct access from VS Code Not checked yet Adds better base branch detection and remembers previous PR creation options for GitHub pull requests Not checked yet Introduces iterative /fix command improvements in the Preview GitHub Copilot UX Quick Chat Not checked yet
9 details in Everything
VS Code 1.80 adds accessible View for screen readers, terminal image support, and WSL WSL Windows Subsystem for Linux, a Microsoft-built compatibility layer that lets a Linux environment run directly on Windows without a virtual machine, giving cyber tools access to Linux binaries, filesystems, and syscalls on a Windows host. over Tunnels.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.80.0 https://github.com/microsoft/vscode.git
$ git checkout 1.80.0
Adds Accessible View mode to improve screen reader support across the editor interface Not checked yet Adds copilot.audio.cues setting to enable audio feedback for GitHub Copilot events Not checked yet Adds terminal.integrated.enableImages config key to allow rendering images directly in the integrated terminal Not checked yet Adds remote.tunnels.allowedPorts configuration section to manage port forwarding rules for remote WSL connections via Tunnels Not checked yet
4 details in Everything
Adds read-only workspace mode, Paste as formatting options, default Git branch setting, notebook rich content search, and linked editing for JSX tags.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.79.0 https://github.com/microsoft/vscode.git
$ git checkout 1.79.0
Adds readonly file attribute to mark specific files and folders in a workspace as read-only. Not checked yet Adds Paste as options to control how item links are formatted when pasted into the editor. Not checked yet Adds automatic copying of external media files into the workspace on drag or paste for Markdown. Not checked yet Adds git.defaultBranchName user setting to set main or a custom branch name for new Git repositories. Not checked yet Enables rich content search in Notebooks, allowing searches based on notebook output and filtering by cell type. Not checked yet
+1 more
show less
Adds linked editing for JSX tags to simultaneously update opening and closing tags. Not checked yet
6 details in Everything
VS Code 1.78 adds drop selector for drag-and-drop links, profile templates, and standalone color picker.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.78.0 https://github.com/microsoft/vscode.git
$ git checkout 1.78.0
Adds DropSelector UI to control how dragged items (links, files) are inserted into the editor Not checked yet Adds profile templates for Python, Java, Data Science to streamline workspace setup Not checked yet Adds standalone color picker dialog for inserting or modifying color formats in text Not checked yet
3 details in Everything
VS Code 1.77 adds notebook format-on-save, Python move symbol refactoring, expanded GitHub Copilot inline/full chat views, and switch-case completions.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.77.0 https://github.com/microsoft/vscode.git
$ git checkout 1.77.0
Adds Notebook Format on Save to automatically format notebook cells when the file is saved Not checked yet Adds Python move symbol refactoring capability to move symbols to an existing or new file Not checked yet Adds 'expanded GitHub Copilot integration' with a preview of inline chat and a full AI chat view Not checked yet Adds TS/JS switch case completions for TypeScript and JavaScript to quickly fill in switch statements Not checked yet
4 details in Everything
VS Code 1.76 introduces Profiles UI, moveable Explorer, notebook kernel MRU, and pytest IntelliSense
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.76.0 https://github.com/microsoft/vscode.git
$ git checkout 1.76.0
Active profile badge and quick switch via Command Palette for managing multiple configuration sets. Not checked yet Move the Explorer view to the secondary side bar or a panel. Not checked yet Notebook kernel picker defaults to Most Recently Used (MRU) list to quickly select recent kernels. Not checked yet Markdown header link suggestions allow linking to headers in files across your workspace. Not checked yet Jupyter IPyWidgets 8 support enables use of the latest IPyWidgets version in notebooks. Not checked yet
+1 more
show less
pytest IntelliSense adds completions for fixtures and parameterized arguments in Python. Not checked yet
6 details in Everything
VS Code 1.75 adds Profiles for sharing settings and VS Marketplace code signing
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.75.0 https://github.com/microsoft/vscode.git
$ git checkout 1.75.0
└──▷ HOW TO FIND IT
Share a custom development environment profile with team members.
📍 Code › Profiles › Create Profile...
Adds Profiles to create and share extension configurations, settings, and keybindings across environments Not checked yet Enables VS Marketplace extension code signing by default for all published extensions Not checked yet
2 details in Everything
VS Code 1.74 adds Remote Tunnels, unsafe repo management, notebook audio cues, and merge editor undo/redo
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.74.0 https://github.com/microsoft/vscode.git
$ git checkout 1.74.0
└──▷ HOW TO FIND IT
Connect to a remote machine without configuring SSH keys or ports
📍 Command Palette › Remote Tunnels: Create Tunnel
Prevent accidental Git operations on shared or external folders
json
copy
Settings > 'git.manageUnsafeRepositories': true
Enable audio feedback during notebook execution and diff review
json
copy
Settings > 'notebook.audioCues': true
Adds Remote-Tunnels to connect to any machine without SSH. Not checked yet Adds ability to manage unsafe Git repositories to prevent operations on folders not owned by the user. Not checked yet Adds audio cues for Jupyter notebook execution results and diff view line changes. Not checked yet Enables JavaScript console.profile collection via the Debug Console. Not checked yet Adds Go to Definition from return values in JavaScript/TypeScript. Not checked yet
+4 more
show less
Adds Just My Code debugging for Python in Jupyter notebooks. Not checked yet Allows requesting GPU GPU A graphics processing unit, a processor that performs many calculations in parallel. AI workloads use GPUs to speed up model training and inference. resources when creating a Dev Container Dev Container A Docker-based development environment defined in a JSON config file, standardized by Microsoft, that lets a project ship its exact toolchain so any contributor's editor or CI system gets an identical workspace. . Not checked yet Allows customizing Explorer auto-reveal logic to control which files scroll into view. Not checked yet Allows hiding Activity bar and Panel badges per view container. Not checked yet
9 details in Everything
VS Code 1.73 adds Dev Container Dev Container A Docker-based development environment defined in a JSON config file, standardized by Microsoft, that lets a project ship its exact toolchain so any contributor's editor or CI system gets an identical workspace. Templates, Search folder filtering, Command Center shortcuts, and more
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.73.0 https://github.com/microsoft/vscode.git
$ git checkout 1.73.0
└──▷ HOW TO FIND IT
Create a Dev Container from a template when starting remote development
📍 Open the Command Palette (Ctrl+Shift+P / Cmd+Shift+P ) and run Dev Containers: Create Dev Container from Template .
Configure search folders to include/exclude specific directories
📍 In the Search view tree, right-click a folder and select Add to Include/Exclude List .
View common modes in Command Center without navigating menus
📍 Hover over or click the Command Center bar at the top of the window to see mode shortcuts.
vscode.dev protected branch workflow creates a new branch when committing to a protected one Not checked yet Dev Container Templates let you create new containers from existing templates Not checked yet New standalone extensions for isort, Pylint, and Flake8 are available for Python Not checked yet Search include/exclude folders can be set quickly in the tree view Not checked yet Command Center mode shortcuts display common modes before recent files Not checked yet
+3 more
show less
Merge editor supports the Combination option to auto-merge conflicts Not checked yet Markdown links update automatically on file or image rename/move Not checked yet New audio cues indicate task completion/failure and Terminal Quick Fix availability Not checked yet
8 details in Everything
VS Code 1.72 introduces nested Git repo detection, Terminal Quick Fixes, Markdown link validation, and Dev Container Dev Container A Docker-based development environment defined in a JSON config file, standardized by Microsoft, that lets a project ship its exact toolchain so any contributor's editor or CI system gets an identical workspace. Features.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.72.0 https://github.com/microsoft/vscode.git
$ git checkout 1.72.0
Adds support for detecting and displaying nested Git submodules in the source control view Not checked yet Adds automatic header, file, and image link validation in Markdown files Not checked yet Adds Terminal Quick Fixes to suggest corrections for command typos and upstream remote setup Not checked yet Adds ability to pin frequently used tasks to the top of the Run Task dropdown via UI Not checked yet Improves Dev Containers Dev Containers A Docker-based development environment specification, maintained by Microsoft and the open-source community, that lets a tool ship a fully configured, reproducible workspace any developer can launch inside VS Code or a compatible editor. experience with new Features management for adding and sharing container functionality Not checked yet
5 details in Everything
Visual Studio Code 1.71 adds merge editor improvements, fish/Git Bash terminal integration, and image pasting in Jupyter notebooks.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.71.0 https://github.com/microsoft/vscode.git
$ git checkout 1.71.0
terminal.integrated.shellIntegration.args now supports fish and Git Bash shells for enhanced shell integration. Not checked yet New smooth scrolling behavior added to the terminal UI. Not checked yet Pressing F2 in the Explorer now offers selection options for filename, whole name, or file extension via File: Rename Selection. Not checked yet Jupyter notebooks now support pasting image files directly into Markdown cells with inline preview. Not checked yet Live Preview extension now supports multi-root web projects. Not checked yet
+1 more
show less
Code Action UI has been redesigned for easier discovery of available code actions. Not checked yet
6 details in Everything
VS Code 1.70 adds sticky scroll, folded selection, tree view search, and --merge CLI flag.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.70.0 https://github.com/microsoft/vscode.git
$ git checkout 1.70.0
Adds --merge command-line option to use the 3-way merge editor as your default merge tool Not checked yet Adds fold selection feature allowing users to create custom folded regions in the editor Not checked yet Adds search multi-select capability to select and act on multiple search results simultaneously Not checked yet Adds tree view search and filtering (tree-find-control) for views like the Find Explorer Not checked yet Adds sticky scroll preview to the editor UI to show current source code scope while scrolling Not checked yet
5 details in Everything
VS Code 1.69 adds a 3-way merge editor, Command Center UI, and terminal shell integration.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.69.0 https://github.com/microsoft/vscode.git
$ git checkout 1.69.0
Introduces the Command Center UI for searching files, running commands, and navigating cursor history Not checked yet Adds Do Not Disturb mode to silence non-critical notification popups Not checked yet Enables quick switching between light and dark themes via a new toggle Not checked yet Adds Terminal shell integration to display command status and run recent commands Not checked yet Highlights task success or failure exit codes with Task output decorations Not checked yet
+4 more
show less
Allows configuring the default Git Commit action with a new Git Commit action button Not checked yet Adds Debug Step Into Target support to step directly into functions when paused Not checked yet Supports JavaScript sourcemap toggling to switch debugging between compiled and source code Not checked yet Adds Color theme tester support for built-in themes in vscode.dev Not checked yet
9 details in Everything
VS Code 1.68 adds .gitignore-based file hiding in Explorer, table view for Problems panel, and Markdown link validation preview.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.68.0 https://github.com/microsoft/vscode.git
$ git checkout 1.68.0
└──▷ HOW TO FIND IT
Clean up the Explorer view by ignoring build artifacts or node_modules using your workspace's existing gitignore rules.
📍 "# In Settings JSON
cmd+k cmd+s
-- add to settings.json: --
explorer.ignoreFiles: [
"/.git/",
"/node_modules/**"
]"
Filter the Problems panel by file path or severity using the new table interface instead of a flat list.
📍 1. Open Problems panel (Ctrl+Shift+M / Cmd+Shift+M)
2. Click the Table view icon in the panel title bar
Hide files in the Explorer using existing .gitignore patterns via the explorer.ignoreFiles setting Not checked yet View errors and warnings in the Problems panel as a sortable table by clicking the table view toggle button Not checked yet
2 details in Everything
Explorer file nesting, Settings editor filters, default bracket pair colorization, toggle inlay hints with Ctrl+Alt, Markdown drag-and-drop links, and Find All References for Markdown.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.67.0 https://github.com/microsoft/vscode.git
$ git checkout 1.67.0
Adds Explorer file nesting to visually group generated files under their source files Not checked yet Adds a filter button to the Settings editor that supports searches like @Modified settings Not checked yet Enables colored bracket pair colorization by default in the editor Not checked yet Adds Ctrl+Alt shortcut to quickly hide or show inlay hints in the editor Not checked yet Allows dragging and dropping files into the editor to automatically create Markdown links Not checked yet
+1 more
show less
Enables Find All References for headers, files, and URLs within Markdown documents Not checked yet
6 details in Everything
Visual Studio Code 1.66 adds Local History, CSS/LESS/SCSS formatters, JavaScript heap profiles, and R language support.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.66.0 https://github.com/microsoft/vscode.git
$ git checkout 1.66.0
└──▷ HOW TO FIND IT
View local file history to restore previous versions of a file without using source control.
📍 Open Command Palette (Ctrl+Shift+P) and search for Local History: Show History or navigate via the Explorer view's context menu on a file.
Format CSS, LESS, or SCSS files in-place using the new built-in formatter instead of an external extension.
📍 Open a .css, .less, or .scss file, then use Format Document (Shift+Alt+F) to apply syntax-aware formatting.
Capture JavaScript memory allocation profiles while debugging an application to identify memory leaks.
📍 Start a debug session, open the Run and Debug panel, expand the JavaScript Heap Profiles section, and click Take Heap Snapshot .
Adds Local History to track file changes independent of source control Not checked yet Adds built-in formatters for CSS, LESS, and SCSS languages Not checked yet Adds JavaScript heap profiling capability during debugging sessions Not checked yet Adds a new topic for R language data science usage in the documentation Not checked yet
4 details in Everything
VS Code 1.65 adds scoped editor history, lazy variable evaluation, terminal shell integration preview, and light high contrast theme.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.65.0 https://github.com/microsoft/vscode.git
$ git checkout 1.65.0
Scope Go Back/Go Forward history to the current editor group or a single editor Not checked yet Add light-high-contrast Light High Contrast theme for enhanced editor visibility Not checked yet Add audio cues for warnings, inline suggestions, and breakpoint hits Not checked yet Enable drag-and-drop of Problems and Search results to new or existing editor groups Not checked yet Automatically close diff editors after Git operations via Source Control diff editor management Not checked yet
+2 more
show less
Add lazy evaluation of JavaScript/TypeScript property getters in the debugger Not checked yet Preview Terminal shell integration for rerunning terminal commands with status decorations Not checked yet
7 details in Everything
New Side Panel, Settings search prioritization, audio cues, binary debug view, notebook output search, and Markdown path suggestions
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.64.0 https://github.com/microsoft/vscode.git
$ git checkout 1.64.0
Adds a Side Panel that displays multiple views simultaneously alongside the main editor area Not checked yet Improves Settings editor search to prioritize whole-word matches over partial matches Not checked yet Adds Audio cues configuration keys to hear alerts when the cursor moves to folded regions, errors, and breakpoints Not checked yet Adds Unicode highlighting improvements to avoid highlighting certain characters in supported languages Not checked yet Adds Automatic terminal replies feature to create automatic responses to common terminal prompts Not checked yet
+5 more
show less
Improves Notebook UI by enabling search within Markdown and output cells Not checked yet Adds a Debug binary data view for viewing and editing binary data while debugging Not checked yet Adds Markdown path suggestions (IntelliSense) to quickly insert relative file paths and header links Not checked yet Adds JS/TS support for surround with snippets to insert selected code inside snippets Not checked yet Adds support for signed GitHub commits in vscode.dev and github.dev environments Not checked yet
10 details in Everything
VS Code 1.63 adds Marketplace theme preview, Screencast Mode command names, and Remote - SSH Reopen in Container command
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.63.0 https://github.com/microsoft/vscode.git
$ git checkout 1.63.0
Adds Marketplace theme preview to try Color Themes without installing them Not checked yet Configures Problems navigation order by severity or file position via problems.view.mode Not checked yet Shows commands in Screencast Mode alongside keyboard shortcuts Not checked yet Improves Notebook Markdown font size adjustment and file/URL linking support Not checked yet Adds Invisible Unicode character highlighting for confusable glyphs Not checked yet
+5 more
show less
Provides TypeScript method signature completions for interfaces and overrides Not checked yet Enables installation of Pre-release extensions from the Marketplace Not checked yet Introduces a new Java welcome experience to help configure and learn about Java Not checked yet Adds more extensions available for vscode.dev Not checked yet Adds Remote-SSH: Reopen in Container command to work in a container on an SSH server Not checked yet
10 details in Everything
Adds preview of vscode.dev and performance improvements for large workspaces.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.62.0 https://github.com/microsoft/vscode.git
$ git checkout 1.62.0
Adds vscode.dev as a new browser-based cyber tooling environment Not checked yet
1 detail in Everything
VS Code 1.61 adds locked editor groups, Jupyter table of contents, and platform-specific extension support
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.61.0 https://github.com/microsoft/vscode.git
$ git checkout 1.61.0
└──▷ HOW TO FIND IT
Lock a terminal group so users cannot accidentally switch it to a file editor.
📍 Use the Locked Editor Groups command in the Command Palette to define which editor types are fixed.
Navigate quickly through large Jupyter Notebooks using the new Table of Contents sidebar.
📍 Open a .ipynb file and locate the Table of Contents view in the sidebar to jump between sections.
Locked editor groups feature lets you select specific editor types to auto-lock within a group, preventing accidental swaps. Not checked yet Jupyter Notebook support gains a Table of Contents view for easier navigation within notebooks. Not checked yet New platform-specific extensions capability allows publishing distinct extension versions for Windows, macOS, and Linux. Not checked yet
3 details in Everything
VS Code 1.60 adds automatic language detection on paste, locked editor groups, and settable watch values.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.60.0 https://github.com/microsoft/vscode.git
$ git checkout 1.60.0
Adds automatic programming language detection when pasting text into the editor Not checked yet Introduces high-performance bracket pair colorization for large files Not checked yet Allows setting debugger watch values directly during a session Not checked yet Adds the ability to lock an editor group to preserve layout Not checked yet Enables rich syntax highlighting for code blocks in the Settings editor Not checked yet
5 details in Everything
VS Code 1.59 adds Jupyter notebook support, Dev Container Dev Container A Docker-based development environment defined in a JSON config file, standardized by Microsoft, that lets a project ship its exact toolchain so any contributor's editor or CI system gets an identical workspace. CLI, testing API, and disassembly view
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.59.0 https://github.com/microsoft/vscode.git
$ git checkout 1.59.0
Adds devcontainer CLI to manage development containers from the command line. Not checked yet Adds built-in .ipynb (Jupyter Notebook) file support for opening notebooks directly in VS Code. Not checked yet Introduces the final Testing API with a native Test Explorer view. Not checked yet Adds preview mode for the Debug Disassembly view displaying disassembled C++ code. Not checked yet
4 details in Everything
Visual Studio Code 1.58 adds terminals in editor area, workspace trust, and Jupyter integration
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.58.0 https://github.com/microsoft/vscode.git
$ git checkout 1.58.0
Adds Terminals in the editor allowing users to create or move terminal instances directly into the editor area Not checked yet Introduces Workspace Trust controls with options to customize settings and disable the Restricted Mode banner Not checked yet Enables Markdown preview KaTeX support for rendering math formulas within markdown files Not checked yet Adds Jupyter Interactive window to execute Jupyter code interactively inside VS Code Not checked yet Adds Jupyter notebook debugging capability allowing users to debug Jupyter notebook code from within the IDE Not checked yet
5 details in Everything
VS Code 1.57 introduces Workspace Trust, Remote Repositories, Notebook API, and Edge debugging
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.57.0 https://github.com/microsoft/vscode.git
$ git checkout 1.57.0
Adds Workspace Trust to restrict code execution in untrusted workspaces Not checked yet Adds support for browsing and editing code in remote repositories without cloning locally Not checked yet Finalizes the Notebook API enabling native notebook extensions Not checked yet Integrates Microsoft Edge Developer Tools into VS Code for browser debugging Not checked yet Adds terminal tabs to manage multiple open terminals within a single window Not checked yet
+1 more
show less
Enables Go to Definition for non-code files such as images and stylesheets Not checked yet
6 details in Everything
Visual Studio Code v1.56 adds terminal tabs preview, inline debug values, and custom default terminal profiles.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.56.0 https://github.com/microsoft/vscode.git
$ git checkout 1.56.0
Adds terminal.tabs view for managing open terminal sessions Not checked yet Enables editor.inlineValues to display variable values during debugging sessions Not checked yet Adds terminal.defaultProfile setting to create a custom default terminal profile Not checked yet
3 details in Everything
Visual Studio Code 1.55 adds terminal profiles, inline breakpoints, and multi-cursor accessibility.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.55.0 https://github.com/microsoft/vscode.git
$ git checkout 1.55.0
Adds terminal.integrated.profiles configuration key to define named shell profiles for the integrated terminal Not checked yet Introduces inline breakpoint menu in the editor gutter for setting conditional/logpoint breakpoints without opening the context menu Not checked yet
2 details in Everything
Adds Apple Silicon support, persistent terminal processes on reload, and Product Icon Themes.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.54.0 https://github.com/microsoft/vscode.git
$ git checkout 1.54.0
└──▷ HOW TO FIND IT
Apply a custom icon theme to change the look of menus and sidebar icons.
📍 Ctrl+K Ctrl+T (or Cmd+K Cmd+T on macOS) and select Product Icon Theme
Apple Silicon builds are now available in the stable release Not checked yet Product Icon Themes allow personalization of VS Code icon imagery Not checked yet Remote ports are displayed in a table widget instead of a list Not checked yet
3 details in Everything
VS Code 1.53.0 adds wrap tabs, tab decorations, conditional exception breakpoints, and Notebook outlines.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.53.0 https://github.com/microsoft/vscode.git
$ git checkout 1.53.0
editor.tabSizing setting supports wrap to replace the horizontal scrollbar with wrapped tabs Not checked yet search.defaultViewMode setting controls whether the Search view or a new Search editor opens by default Not checked yet Node.js worker_threads debugging support enables debugging of Worker code alongside main process Not checked yet Outline view for Notebook cells appears in the sidebar to improve navigation within notebooks Not checked yet Markdown preview automatically updates when images are changed on disk without manual refresh Not checked yet
5 details in Everything
VS Code 1.52 adds extension bisect, word wrap in diff editor, undo file ops, terminal config menu, and new Git settings.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.52.0 https://github.com/microsoft/vscode.git
$ git checkout 1.52.0
└──▷ HOW TO FIND IT
Launch the extension bisect workflow from the Command Palette
📍 >Extension: Start Extension Bisect
terminalDropdown.menu.enabled allows modifying terminal settings directly from the terminal dropdown menu Not checked yet New extension bisect feature to identify problematic extensions Not checked yet
2 details in Everything
VS Code 1.50.0 adds Webview Views API, Linux ARM ARM A family of RISC processor architectures designed by Arm Holdings and licensed to chip makers worldwide; tools target it to run natively on mobile, embedded, and increasingly server and desktop hardware. support, and settings editor accessibility improvements.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.50.0 https://github.com/microsoft/vscode.git
$ git checkout 1.50.0
Adds webviewViews contribution point to allow extensions to register views in the sidebar and panel Not checked yet Adds official support for Linux ARMv7 and ARM64 ARM64 A 64-bit CPU architecture from ARM Holdings, used in mobile chips, Apple Silicon, and cloud servers; cyber tools target it to run natively on those platforms instead of relying on emulation. architectures Not checked yet Adds keyboard navigation capabilities to the Settings editor for improved accessibility Not checked yet
3 details in Everything
VS Code 1.49 adds format-on-change filtering, search casing toggles, multi-repo source control view, and debug console filters.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.49.0 https://github.com/microsoft/vscode.git
$ git checkout 1.49.0
Format on Save now supports a formatOnSaveMode option to limit formatting to only the modified text in the editor Not checked yet Source Control view gains a nested Repositories section displaying pending changes grouped by repository when multiple repos are open Not checked yet
2 details in Everything
VS Code 1.48 adds Settings Sync to stable, browser debug without launch.json, and public repo publishing
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.48.0 https://github.com/microsoft/vscode.git
$ git checkout 1.48.0
└──▷ HOW TO FIND IT
Start debugging a browser page without configuring a launch file.
📍 Debug: Open In Browser
Push the current workspace to GitHub and choose visibility.
📍 GitHub: Publish to Repository...
Brings Settings Sync to the stable channel for cross-machine config backup Not checked yet Enables in-browser debugging via the Debug: Open In Browser command without a launch.json Not checked yet Adds GitHub: Publish to Repository... with options for public or private repositories Not checked yet
3 details in Everything
VS Code 1.47 introduces stable Windows ARM ARM A family of RISC processor architectures designed by Arm Holdings and licensed to chip makers worldwide; tools target it to run natively on mobile, embedded, and increasingly server and desktop hardware. builds, a single-view source control interface, and hex file editing via HexEditor extension.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.47.0 https://github.com/microsoft/vscode.git
$ git checkout 1.47.0
Adds stable availability for Windows on ARM architecture Not checked yet Introduces a unified Source Control view displaying all repositories in a single UI pane Not checked yet Enables sorting pending changes by name, path, or status within the unified source control tree Not checked yet Allows direct editing of complex object settings from the Settings editor UI without opening raw JSON files Not checked yet Adds HexEditor extension support for viewing and editing files in hexadecimal format Not checked yet
5 details in Everything
VS Code 1.46 adds flexible layout dragging, pinned tabs, GitHub issue notebooks, and Windows ARM64 ARM64 A 64-bit CPU architecture from ARM Holdings, used in mobile chips, Apple Silicon, and cloud servers; cyber tools target it to run natively on those platforms instead of relying on emulation. builds.
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.46.0 https://github.com/microsoft/vscode.git
$ git checkout 1.46.0
└──▷ HOW TO FIND IT
Run a query for open GitHub issues and view the results in a notebook cell.
📍 Create a new Notebook file (.ipynb) and add a GitHub Issue Query cell to execute your search.
Adds ability to move and group views in the Side Bar and Panel for Flexible view and panel layout Not checked yet Pins editors as tabs in the title bar via Pinned editor tabs to keep frequently used files handy Not checked yet Introduces GitHub Issue Notebooks allowing users to run GitHub issue queries and display results in a custom Notebook Not checked yet Adds Windows ARM64 support with new VS Code Insiders builds Not checked yet
4 details in Everything
VSCode 1.45 adds WSL WSL Windows Subsystem for Linux, a Microsoft-built compatibility layer that lets a Linux environment run directly on Windows without a virtual machine, giving cyber tools access to Linux binaries, filesystems, and syscalls on a Windows host. 2 remote support, GitHub auth, semantic token styling, and mouse wheel tab switching
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.45.0 https://github.com/microsoft/vscode.git
$ git checkout 1.45.0
Adds automatic GitHub authentication for repository links, enabling seamless access to Pull Requests and Issues without manual token configuration Not checked yet Enables inline hovers and suggestions for GitHub issues and users directly in the editor via the GitHub API integration Not checked yet Adds editor.tokenColorCustomizations settings section allowing users to define custom colors for semantic tokens (e.g., function calls, variables) Not checked yet
3 details in Everything
VS Code 1.39 adds Source Control tree view, WSL WSL Windows Subsystem for Linux, a Microsoft-built compatibility layer that lets a Linux environment run directly on Windows without a virtual machine, giving cyber tools access to Linux binaries, filesystems, and syscalls on a Windows host. container remote explorer, and custom terminal working directory shortcuts
└──▷ GET THIS VERSION
copy
$ git clone --branch 1.39.0 https://github.com/microsoft/vscode.git
$ git checkout 1.39.0
Adds a Source Control: Change in Diff command (and corresponding tree-view UI) to display pending changes as a tree alongside the existing list view. Not checked yet Adds Remote Explorer support for displaying WSL distributions and repository containers in the sidebar. Not checked yet
2 details in Everything