Skip to main content
AI Development

VS Code Local harness ends February 2027: what to move first

VS Code 1.146 deprecates the Local harness for Copilot agent sessions. Removal follows in February 2027. Prompt files, hooks and MCP secrets need manual moves.

Sean McLellan profile photo

Sean McLellan

Lead Architect & Founder

7 min read
Flat illustration on a cream background of an old, open, nearly empty wooden toolbox on the left and a newer copper-trimmed wooden toolbox on the right holding small brass wrenches, a coil of copper wire, a valve and a hook, with a plank between them carrying three brass fittings and a blank tag
Constructed diagramIllustration: fittings carried from an old toolbox to a new one, for customizations moving from the Local harness to the Copilot harness. Not product UI.

Visual Studio Code 1.146 deprecates the Local harness, the older runtime behind GitHub Copilot agent sessions in the editor, and VS Code will remove it in February 2027. GitHub announced the change in a November 16 changelog post. Its replacement, the Copilot harness, has been the default for new sessions since VS Code 1.143. The temporary admin setting that keeps teams on the old runtime goes away at the same time. Teams still on the Local harness have about three months to move their prompt files, hooks and MCP servers.

Picking the new harness is one menu choice. The work is in the customizations and controls that don't come along with it.

What is the Local harness, and what replaces it?

A harness is the software that sits between the model and your project. It prepares the context the model sees, runs tool calls, handles approvals and keeps the conversation going. The VS Code transition guide calls the Local harness "the legacy, editor-specific implementation." The Copilot harness is built on the Copilot SDK and shares its coding-agent runtime with Copilot CLI and the GitHub Copilot app.

Some points GitHub and the VS Code docs make clear:

  • Your files stay where they are. GitHub's changelog says choosing the Copilot harness "doesn't move your files to the cloud." The harness docs add that "Local is a harness name, not a synonym for running on your machine."
  • Your plan doesn't change. If you reach models through GitHub Copilot, you keep your existing plan, including Copilot Free, according to the changelog.
  • Nothing extra to install. VS Code ships the runtime. You don't need Copilot CLI or the SDK on the machine.
  • Feature work focuses on the Copilot harness. The Local harness stays available during the transition where policy permits, but GitHub says new features and improvements will focus on the Copilot harness.

The harness controls what context and tools the model gets, and the guide warns: "The same prompt and model can therefore produce different results in Local and Copilot harness sessions." BaristaLabs did not run either harness for this post. The checks below come from GitHub's and VS Code's documentation.

What doesn't carry over when you switch?

Switching harnesses doesn't automatically convert every customization. The guide's checklist covers each item:

Comparison table titled Moving from the Local harness to the Copilot harness in VS Code, with columns Item, What carries over and What to do. Prompt files: not converted automatically; convert to skills and recheck agent, model and tools. Hooks: no automatic conversion; review events, inputs, outputs and failures. MCP servers: input prompts and stored secrets don't transfer; re-verify auth, tools and sandbox behavior. Personal agents and instructions: files in editor profile storage need to move; use the Copilot harness native locations. Extension tools: need a connected editor window; check tools in a Copilot session. Local memory files: not converted to Copilot Memory; keep team guidance in instructions or skills. The setting vscode.temporaryDefaultToLocalHarness: removed with the Local harness; remove the exception before Feb. 2027.
Constructed diagramConstructed diagram based on the VS Code transition guide and GitHub's November 16 changelog (November 2026). Not product UI.
  • Prompt files. Copilot harness sessions "do not load *.prompt.md files," according to the customization migration docs. The built-in conversion to skills keeps each prompt's name, description and argument guidance. It drops any agent, model or tool selection the prompt set, so check those by hand.
  • Hooks. "Hooks have no automatic conversion," the guide says. Review the events each hook listens for and what it does when it fails.
  • MCP servers. Input prompts and stored editor secrets "don't automatically transfer." The guide also says not to replace a secret placeholder with a literal credential while you fix this.
  • Personal agents, instructions and skills. Files in editor profile storage need to move to the harness's native folders, such as .github/agents, .github/instructions and .github/skills in a repository, or ~/.copilot/skills for a user.
  • Extension tools. These still need an editor window where the extension is running.
  • Memory. Local harness memory files and Copilot Memory are separate systems, and switching doesn't convert the files. Copilot harness sessions use cloud-backed Copilot Memory by default, subject to account and organization settings.

To find what needs work, run Chat: Open Customizations and select Migrations. VS Code lists what it detected. Migrated files don't sync with the originals, so keep backups until the new setup works and then remove the duplicates. The experimental Migrate with Agent option uses Copilot credits.

How do developers switch an existing session?

Pick one small task that already works in the Local harness and write down the result you expect. Note the tools, customizations and approvals it needs. Then:

  1. Back up your customization files and settings.
  2. Start a new chat, open Session Target and select Copilot.
  3. To keep an old conversation, open the Local session, open Session Target in its chat input and select Copilot. The guide says the handoff carries the conversation history and context. It adds: "This is not a two-way harness switch."
  4. Run the task and compare the actual edits and test results, not just the agent's summary.

One practical difference: closing the editor window stops a Local harness session's work. A Copilot harness session keeps running while its Agent Host process stays up, and you can continue it in the Agents window.

What should VS Code admins change before February 2027?

BaristaLabs interpretation: the admin decision is less about the deadline than about the controls. GitHub says the Copilot harness can enforce "managed runtime permissions and sandbox restrictions that the Local harness cannot enforce." While a team stays on the Local fallback, those controls don't apply to its sessions.

The guide lays out three settings, and they do different things:

Scroll sideways to see all 3 columns.

SettingWhere it goesWhat it does
ChatEditorPreferCopilotHarness: trueDevice policyPicks the Copilot harness for new editor chats without blocking others
temporaryAllowLocalHarness: falseEnterprise-managed settings (VS Code 1.143+)Prohibits the Local, Claude and Codex harnesses
temporaryDefaultToLocalHarness: trueEnterprise-managed settings (VS Code 1.142+)Temporary exception that keeps Local available; removed in February 2027

Two details are easy to miss. First, temporaryAllowLocalHarness: false blocks the Claude and Codex harnesses along with Local, so check whether anyone relies on them before you deploy it. Second, both keys belong in enterprise-managed settings, not in user or workspace settings.

Requiring the Copilot harness doesn't turn on sandboxing or add permission rules. You deploy those separately. Since VS Code 1.143, a nonempty managed permissions.deny list or sandbox.enabled: true also requires the Copilot harness, unless the temporary Local default applies.

If you set the exception, know what it costs. The guide warns that falling back to the Local harness "does not preserve those protections." It also says: "An administrator must review and accept this gap before permitting fallback." GitHub's changelog adds that leaving the exception configured "won't keep the Local harness available" once it's removed.

How do you verify the rollout?

The guide suggests a pilot that covers your real projects and execution environments, including Remote SSH, WSL and Dev Containers where your teams use them. After deploying the settings:

  • Start a new editor chat and a new Agents window session, and confirm both use the Copilot harness.
  • Run Developer: Policy Diagnostics to check the effective policy values and where each came from.
  • In a Copilot session, run /sandbox policy to check the execution host, sandbox state, and file system and network limits.
  • Test one approved operation and one harmless operation that policy should block, and confirm the block happens. An approval prompt on its own doesn't count.
  • Repeat on remote hosts and containers, because the guide says to check the machine where commands run, not only the connecting device.

OpenTelemetry export works for both harnesses, so existing agent dashboards shouldn't force anyone to stay on Local. Check the exporter destination and attributes after the switch anyway.

Recommendation

BaristaLabs recommends these steps, in this order:

  1. Set ChatEditorPreferCopilotHarness now. It moves new editor chats to the Copilot harness without prohibiting the others.
  2. Have each team work through the Migrations list by the end of 2026: convert prompt files to skills and review hooks by hand.
  3. Pilot the managed sandbox and deny rules on the Copilot harness, with the block test above.
  4. Only then set temporaryAllowLocalHarness: false, after confirming nobody depends on the Claude or Codex harnesses.

Use temporaryDefaultToLocalHarness only for a named integration with an owner and a removal date before February, and record that the security controls don't apply while it's on.

If you're also setting model defaults for the new harness, see our notes on making Copilot Fusion the default and on putting the Copilot model-policy API under change control. For JetBrains users on the same managed settings, see running a governed Copilot pilot in JetBrains and testing what a managed sandbox actually blocks. For help with an agent rollout plan, see BaristaLabs AI consulting.

Sources

Copilot harness migration

Move your teams off the Local harness before February

Share your VS Code managed settings and the customizations your teams rely on. BaristaLabs can help sort what converts, what needs a manual rewrite, and which sandbox and permission rules to pilot before you require the Copilot harness.

A copy of your managed settings and a list of prompt files and hooks is enough to start.

Turn this idea into a pilot

Which workflow should go first?

Use the readiness check to compare impact, effort, risk, owner, and next step before requesting a review.

  • 3-5 minutes
  • Deterministic score
  • No sensitive data
Check workflow readiness

Practical AI Workflow Notes

Want more practical AI operations ideas?

Get short notes on applying AI inside real small-business workflows — from document handling and customer follow-up to internal reporting, compliance, and automation guardrails.

A useful next step if you’re still exploring and not ready to request a 20-minute workflow assessment.

Occasional emails. Practical workflow guidance only. Unsubscribe anytime.