Visual Studio Code and GitHub Copilot - What's new in 1.109.3 and 1.110

Fokko Veegens covers the highlights of VS Code 1.109.3 and 1.110, including agent hooks for deterministic control, mid-session message steering and queueing, skills as slash commands, the new agent debug panel with flowchart visualization, edit and ask mode changes, agent plugins from the marketplace, #usages and #rename refactoring tools, and AI co-author attribution for commits.

Full summary based on transcript

Visual Studio Code versions 1.109.3 and 1.110 bring a wide range of improvements to GitHub Copilot, from deterministic agent control through hooks to a greatly improved debugging experience and a new plugin ecosystem. This highlights video covers the key features Fokko considers most relevant.

Agent Hooks

Hooks give you deterministic control over Copilot by running scripts at specific points during a session.

Available hook types:

• Session start - triggered when a new agent session begins
• Pre-tool use - triggered right before Copilot uses a tool (e.g., the terminal tool)

Setup:

• Type /hooks in chat to see all available hooks and create a new one
• Select the hook type and provide a file name
• A hook file is created in the .github/hooks/ folder

Hook file format:

• The hook file specifies a command to execute (e.g., a PowerShell script)
• The command type can vary per platform (Linux, Windows, macOS)
• The script can return a JSON response to Copilot with:
  • continue - whether Copilot should continue execution (true or false)
  • stopReason - message displayed in chat when execution is stopped
  • systemMessage - additional context for Copilot

Example use cases:

• Setting a proxy before tool execution
• Running linters after Copilot makes suggestions
• Filtering or blocking specific tool calls
• Enforcing policies before terminal commands

Configuration:

• Setting: chat.hooks.enabled - must be enabled for hooks to work
• Hooks in the project .github folder take precedence over user profile hooks
• Enterprise policies can disable hooks entirely

Message Steering and Queueing

Mid-session steering allows you to send additional instructions to Copilot while it is actively working, a feature previously only available in the coding agent on github.com.

How it works:

• While Copilot is processing a request, type a new message in the chat input
• Three options appear:
  • Steer with message - message is handled after the current tool call completes
  • Add to queue - message is executed once Copilot has fully completed
  • Stop and send - stops current execution and sends your new message

Configuration:

• Setting: chat.requestQueuing.defaultAction
• Default value: steer (pressing Enter sends as a steering message)
• Can be changed to queue (stop and send is not available as a default action)

Skills as Slash Commands

Skills are markdown-based definitions that provide Copilot with specialized behavior and additional tools for deterministic outcomes.

Structure:

• Skills live in a skills/ directory with a subdirectory per skill
• Each skill requires at minimum a skill.md file
• The skill.md has a frontmatter header with name and description

Invocation:

• Copilot can automatically discover and use skills
• Manually invoke with /skillname (e.g., /net for a .NET design pattern skill)
• Ask Copilot “What skills do you have available?” to list all discovered skills

Frontmatter settings:

• userInvocable - set to false to prevent manual slash command invocation (default: true)
• disableModelInvocation - set to true to prevent Copilot from automatically picking up the skill

Resources:

• The “Awesome Copilot” GitHub repository contains a curated collection of skills

Agent Debug Panel

A new structured debug panel replaces the previous markdown-heavy debug view for inspecting Copilot sessions.

Access:

• Open via command palette (F1 / Ctrl+Shift+P) and search for “Agent Debug Panel”

Dashboard view:

• Shows recent sessions with summary stats: model turns, tool calls, total tokens, errors, and total events

Agent flowchart:

• Visual flowchart of the entire session
• Scroll to zoom, drag to pan
• Expandable nodes showing:
  • Loaded instructions, skills, agents, and hooks
  • User messages with prompt sections and attachments
  • Tool calls with details (e.g., which files were read)
  • Model calls with token counts

View logs:

• Alternative structured log view with similar detail
• Useful for finding specific tool calls and their parameters

Edit and Ask Mode Changes

Edit mode has been hidden by default and reimplemented as a custom agent, while ask mode has gained agentic capabilities.

Edit mode:

• Hidden from the mode menu by default
• Re-enable via setting: chat.editMode.hidden (uncheck to show)
• Now implemented as a custom agent definition (no longer a proprietary mode)
• Has a “Configure tools” option - most built-in tools are disabled by default
• Still restricts edits to the active file and explicitly attached files
• View the agent definition via “Configure custom agents” to create your own variants

Ask mode:

• Now has “Configure tools” - full agentic experience with tool access
• Can discover context using tools instead of requiring explicit context attachment
• Also backed by a custom agent definition that you can customize

Agent Plugins

Agent plugins package multiple Copilot customizations (instructions, skills, custom agents) into a single installable unit.

Plugin contents:

• Custom instructions, skills, agents, and other customizations bundled together
• A .github/plugin/ folder with a manifest.json for marketplace metadata
• A README.md explaining usage

Installation:

• Search for “agent plugins” in the VS Code extensions marketplace
• Default plugin sources: github/copilot-plugins and github/awesome-copilot repositories
• First install clones the repository locally in the background
• Also installable via the Copilot CLI (requires separate CLI installation)

Example - Nuke Mode:

• A plugin from the Awesome Copilot repository
• Activate with /noop mode (or the specific slash command for the plugin)
• Changes Copilot's response style: clear step-by-step plans, risk assessments, and structured explanations of changes

Configuration:

• Setting: chat.plugins.enabled - enable/disable the plugin feature
• Setting: chat.plugins.marketplaces - array of GitHub repository references to use as plugin sources
• Push marketplace settings via .vscode/settings.json to share with your team

Tools for Usages and Rename

New #usages and #rename tools improve refactoring workflows in Copilot.

Usage:

• Use #usages to find all usages of a symbol
• Use #rename to rename a symbol across the codebase
• Example prompt: “Find #usages of logger and rename them to classLogger”

Model compatibility:

• Not all models pick up these tools equally well
• Gemini 2.5 Flash showed good results in the demo
• Explicitly mentioning the tool names in your prompt improves reliability

Verification:

• Use the Agent Debug Panel to verify whether the tools were actually invoked
• The flowchart shows exact tool calls: list code usages and rename

AI Co-author Attribution for Commits

A new setting allows adding GitHub Copilot as a co-author on commits made with AI assistance.

Configuration:

• Setting: git.ai.addCoAuthor
• Options:
  • Off (default) - only you appear as author
  • Chat and agent - adds Copilot as co-author when agent or ask mode was used to generate code
  • All - also includes code completions and inline chat

Result:

• Commits show “Co-authored-by: Copilot copilot@github.com” in the commit metadata
• Visible in GitHub commit history alongside your name

Known limitation:

• Does not work when the commit message itself is generated by Copilot (bug as of this release)

Additional Features Mentioned

Several other noteworthy features were briefly mentioned but not demoed in detail:

• Chat conversation forking - fork from a specific point in your chat conversation to a new conversation, carrying along all context
• Copilot memory - Copilot now has memory capabilities, especially useful in plan mode where it retains the generated plan
• Conversation compaction - Copilot automatically summarizes long conversations in the background, and you can now trigger this manually as well