Short answer: Windows 11 coding agent telemetry is real: a Windows 11 system component called MissionControlAggregator records whether you have run specific coding agents, including Claude, Codex, GitHub Copilot, Cursor and Google’s Antigravity, plus common developer tools such as Python, Node.js, GCC and VS Code. It turns that into a small daily report and, if optional diagnostic data is on, queues it for upload to Microsoft. Security researcher Xusheng Li found it, and says optional diagnostics were on by default in the US setup he tested. The report holds on or off flags, not prompts or code. You can stop it in Settings in under a minute.
Key points
- Windows 11 checks process names against a built in list of coding agents and developer tools.
- A daily report says which of them ran. It does not include prompts, source code or how long each tool was used.
- It travels with optional diagnostic data, which Li found switched on by default on a clean US install.
- Microsoft had not commented on the finding when we published, and its public docs do not name this component.
- Turning off Send optional diagnostic data stops the upload.
What did the researcher find?
In a detailed teardown, Li describes asking an AI agent to search recent Windows binaries for undocumented features. One lead stood out: a list of executable names inside MissionControlAggregator.dll, a Windows system file.
The DLL compares the programs you run against named slots:
- Command line agents (4 slots):
copilot.exe,claude.exe,codex.exeandagy.exe, with matching names for programs run inside WSL. - Desktop agents (5 slots):
github.exe,claude.exe,codex.exe,cursor.exeandantigravity.exe. - Developer tools (12 categories): including
node.exe,python.exe,dotnet.exe,gcc.exe,clang.exe,gdb.exe,javac.exe,devenv.exe(Visual Studio) andcode.exe(VS Code).
The agent checks sit behind a feature flag with a very direct name: CLICodingAgentUsage. On Li’s Windows 11 test machine that flag was already enabled as a service experiment, and he had not switched it on himself.
The post reached the Hacker News front page with more than 600 points and nearly 500 comments, and outlets including GIGAZINE have picked it up.
How does Windows 11 coding agent telemetry work?
This Windows 11 coding agent telemetry does not scan your running processes itself. It listens to events Windows already records:
- For normal Windows apps, it reads app state and app interactivity summaries from the kernel and window manager.
- For WSL, it reads the Linux subsystem’s program launch events.
That means a regular Windows install of codex.exe is detected, not only Linux tools running inside WSL.
When a name matches, the component sets a flag in its saved state on disk. Roughly once a day it packs those flags into a report called Segments.SubCohorts, where each agent or tool category is one bit: on if it ran during the window, off if it did not.
Li tested this directly. He copied the Windows command prompt, renamed the copy codex.exe and ran it. The next report flagged Codex as used. A copy renamed gcc.exe set the compiler and debugger flag. So the match is on the file name alone, and some slots (Node.js, Python, .NET, Visual Studio) also check the publisher.
What gets sent, and what does not?
Based on Li’s analysis, the report contains:
- Which of the listed agents and tool categories ran in the reporting window
- The time window the report covers
It does not contain prompts, source code, file names from your projects, or how long you used each tool. The source events carry runtime, focus time and resource use, but Li says this component throws those details away.
That is a narrow payload. It is still a record of your professional toolchain, and Microsoft’s own diagnostics and privacy page says diagnostic data is “stored with one or more unique identifiers that can help us recognize an individual user on an individual device.”
Is it on by default?
The upload only happens when optional diagnostic data is allowed. Li found the agent report in the Windows telemetry service’s upload queue, marked as not required, which places it in the optional tier.
He then did a clean install of Windows 11 Home with the region set to the United States. Before he changed anything, the Diagnostic data toggle was already on and labeled Include Optional. If you clicked through setup with the defaults, assume it is on.
Two honest caveats. Li tested one ARM64 VM and one clean US install, so other regions, editions and managed corporate devices may differ. And Microsoft says it collects some optional data from only a sample of devices, so not every PC with the setting on will necessarily send this exact report.
What has Microsoft said?
Nothing specific so far. As of publishing, Microsoft had not responded publicly to the finding or explained why Windows tracks coding agents. Its general documentation lists app activity as optional diagnostic data, used to understand problems and improve products, but it does not name MissionControlAggregator, the CLICodingAgentUsage flag or this report.
What it means for builders
Our read: this looks like market measurement. Microsoft sells GitHub Copilot and VS Code, and this report would show, across a sampled Windows install base, how often developers run Copilot next to Claude, Codex, Cursor or Antigravity. That is our interpretation, not something Microsoft has confirmed.
Why it matters even without prompts or code:
- Client and NDA work: some contracts limit what tooling data leaves a machine. A toolchain fingerprint you never knowingly agreed to is worth knowing about.
- Company laptops: IT teams that already govern AI tools should decide on purpose whether this data goes out, rather than inheriting a consumer default.
- Trust in defaults: the bigger lesson is that the operating system under your agents is collecting data about them too. Agent security reviews usually stop at the agent’s own telemetry settings. Recent cases like Claude Code saving passwords to GitHub and Postman quietly installing plugins into Claude Code and Codex show the same blind spot.
How to turn it off
On a personal Windows 11 PC:
- Open Settings > Privacy & security > Diagnostics & feedback.
- Turn off Send optional diagnostic data.
- Optional: switch on the Diagnostic Data Viewer on the same page to see what Windows collects. Note that Delete diagnostic data removes diagnostic data for the device (not data tied to your Microsoft account) and does not stop future sending.
Microsoft says your device stays just as secure and works normally with only required data.
For managed fleets, admins can set the diagnostic data level by policy: in Group Policy, go to Computer Configuration > Administrative Templates > Windows Components > Data Collection and Preview Builds > Allow diagnostic data and choose required diagnostic data only. Check your MDM baseline too, since many organizations set this once and never revisit it.
What to watch
- Whether Microsoft explains the feature, documents it, or moves it behind a clearer opt in.
- Whether the list of tracked agents grows as new coding tools ship.
- Whether other operating systems or IDEs are found doing the same kind of toolchain measurement.



