Microsoft Ships VS Code 1.145.1, Fixing Two Telemetry Bugs That Almost Nobody Will Notice
Microsoft has released VS Code 1.145.1, a patch that does very little and does it well. Shipped two days after the base 1.145 update, the patch carries forward the entire 1.145 feature set and adds only two targeted corrections. They both live inside the telemetry that powers Copilot's Agent Host, the background process that keeps AI sessions alive across windows. Both were found and fixed internally by Microsoft's release team.
It's the kind of update that won't show up in any press roundup. But it's also a telling snapshot of where the editor is going. On paper, these are two of the most obscure bug fixes you'll read all week. That's the point.
The first (#346287 / PR #346280) deals with engine snapshot pairing IDs. When an Agent Host snapshot had no input/output pairing ID, the telemetry would silently substitute an unrelated SDK model-call ID. That meant analytics could correlate a model's inputs and outputs under a key that didn't actually belong to that conversation. The fix preserves genuine engine pairing IDs and leaves snapshots without a pairing ID unpaired, while keeping conversation identity intact for analysis. Microsoft says it was verified with 48 regression tests across the router, restricted sender, and Copilot forwarder suites.
The second (#346286 / PR #346275) is about edit-survival account attribution. When edits were reported at commit time, the telemetry was missing the originating account identity, which broke attribution in account-based analytics, including after an account switch. The backport, porting the main fix #345691, keeps the Copilot identity attached at commit time without exposing identity in protocol data or public source keys. It also backports a WSL smoke-setup reliability improvement (#345598), verified against 77 passing tests of the Agent Edit Attribution Service.
Both tell you the same thing: VS Code's agent layer now emits rich analytics about model calls, edits, and account usage, and Microsoft apparently cares that the data is correct.
What Else Came Along
The patch also addresses two things you'd actually notice.
If an organization disabled the Claude or Codex agent, users who had picked that agent were stuck with a read-only chat, no input box, anywhere from New Chat to new windows. That's now #345207, fixed. New chats fall back to the default harness while keeping your unsent prompt, and the previous agent choice is restored once the policy lifts.
The second (#344895) is more interesting than it sounds. When an org set MCP access to "registry only," a failed policy refresh, network error, rate limiting, whatever, would temporarily drop the restriction entirely. VS Code now holds onto the last known registry policy and only clears it when GitHub reports that no registry applies. That closes what was effectively a governance gap, if only a transient one.
The Bigger Story
Here's what I found more striking than the bug fixes themselves. The 1.145.x releases are the latest chapter in VS Code's pivot from a place you type code into a control center for AI agents.
You can trace the arc pretty cleanly. In 1.142, the Copilot harness started becoming the default for new sessions, with Canvases entering Preview. By 1.143 the Copilot harness, powered by the Agent Host and the Agent Host Protocol, or AHP, was the default. And 1.144 through 1.145 have been about refinement: remote agent hosts, cron-based automations, mix-and-match harnesses, verbosity controls, and a batch of enterprise governance tools.
Microsoft is consolidating around a single agent foundation (the Copilot SDK, or the Copilot harness) while keeping the older Local harness around as an option. The 1.145.1 telemetry fixes fit right into that. As agents run more persistently and emit more analytics, the correctness of the underlying data pipeline stops being a niche concern and becomes operationally important.
There's also a bit of contrarian signaling in this release. Microsoft removed several agent tools from the Local harness, Install Extension, Run VS Code Command, and Create New Workspace, nudging users toward "standard" agent workflows. Whether that's the right call depends on how much you like scripting your own agent behavior. The Local harness still exists, but it's getting leaner each cycle.
How the Patch Got Built
The "1.145.1 Endgame" issue (#346279) is a window into Microsoft's release machinery, and there's one detail worth flagging.
Candidate issues were reviewed and assigned to the recovery team, peer-reviewed on main, then cherry-picked onto the release/1.145 branch. Version bumping, the commit 6bab64e2d12b04d04c0d99c27ff166023efd1dda, was tracked under milestone #346279 (milestone #468). Builds ran in Azure DevOps on the Monaco pipeline: Stable build 493338, internal Stable 493439, a superseding rollout 493451, and a website production build 493455.
Notably, the standard 24-hour soak of the Insiders build before release was skipped. It wasn't passed off as an approval, the deviation was just recorded transparently in the checklist. Chat-lib publication was deemed not applicable since only engines.vscode changed, and the vscode.dev deploy step was dropped. Aaron Munger (amunger) led the release, with Deepak Kumar (deepak1556) handling build and release engineering.
As of reporting, 607 commits have landed on main since 1.145.1 shipped, a reminder that this patch is a snapshot in a very active codebase.
Where to Get It
Visual Studio Code 1.145.1 is available now in the Stable channel. Head here to the Visual Studio Code download page.
