Windows 11 coding agent telemetry: what the optional diagnostic data switch stops in MissionControlAggregator, and the developer flag Microsoft lists as required
The per-agent bitmap is tagged optional, so the usual switch should keep it home. A coarser IsDeveloper cohort from the same DLL sits in Microsoft's required event list, which Home and Pro cannot opt out of, and nobody has yet published a run with the switch off.
Contents
On November 16, Xusheng Li published a reverse-engineering writeup of a Windows 11 component called MissionControlAggregator.. It watches for coding agents by process name (Claude Code, Codex, Copilot CLI, Cursor, Antigravity and Google’s agy), plus a list of developer tools, and about once a day it turns what it saw into a small bitmap for Windows diagnostics. The writeup reached the Hacker News front page the same day. I found no public statement from Microsoft as of November 17.
The fix that went around with it is one toggle: turn off optional diagnostic data. By the report’s own isRequired: false tag, that should keep the per-agent bitmap off Microsoft’s servers. Nothing documented says it stops the component, though, and Microsoft’s list of required Windows 11 events, the ones Home and Pro cannot opt out of, already carries a second event from it that classifies the user as a developer.
Nothing here was run on a Windows machine for this article. The findings are Li’s, the steps are Microsoft’s documented ones, and each passage says which.
What does Windows 11’s MissionControlAggregator record about coding agents?
Which agents and tool families ran in a reporting window, one bit each, and nothing about what they did. Li found the checks behind a feature named CLICodingAgentUsage, feature ID 62580080. There are four command-line slots (copilot.exe, claude.exe, codex.exe, agy.exe, also matched without the extension when they run inside WSL), five desktop-app slots (github.exe, claude.exe, codex.exe, cursor.exe, antigravity.exe) and twelve broader tool categories fed by names like node.exe, python.exe, dotnet.exe, gcc.exe, clang.exe, javac.exe, devenv.exe and code.exe.
The agent slots match on the name alone. Li copied cmd.exe to a file called codex.exe, ran it, and the Codex CLI bit was set. Some tool slots also check the publisher (Node.js Foundation for Node, Python Software Foundation for Python, Microsoft for .NET and Visual Studio), but the compiler and debugger group does not: a second copy of cmd.exe, named gcc.exe, set that category’s bit.
It does not hook your processes itself. It reads events Windows already produces: KernelProcess. for command-line programs, Win32kTraceLogging. for desktop apps, and the WSL launcher’s Exec events for Linux binaries. Those source events carry run time, focus time and resource use. Li’s reading is that the aggregator drops all of that and keeps one bit per slot, so the timestamps in the report mark the collection window, not how long you used anything. He found no prompts or code in it.
The report from the codex.exe test, as he pulled it from the upload queue:
{
"name": "Microsoft.Windows.MissionControlAggregator.Segments.SubCohorts",
"ext": { "privacy": { "isRequired": false } },
"data": {
"Version": "0.1.5",
"CliAgentSubCohorts": [4],
"DesktopAgentSubCohorts": [0],
"FlushedFromSave": true
}
}The 4 is 1 << 2, the Codex CLI slot. The state between reports lives on disk:
C:\ProgramData\Microsoft\Diagnosis\AggregatorStorage\MissionControlAggregator$Is the Windows 11 coding agent telemetry on every PC?
Not proven, and Li says so himself. On his Windows 11 ARM64 VM (build 26200.9974, the DLL at 10.0.26100.9818) the feature was already on “as a service experiment”, without him enabling it. On a VM built from the latest public ISO, the feature was off by image default. On the evidence of those two machines, it ships in the binary and Microsoft switches it on per device from its side. Nobody outside Microsoft knows the share.
Whether the report leaves the machine is the second gap. Li found the event in the DiagTrack upload queue with optional data allowed; he did not capture it on the wire, so “queued” is what was observed. Microsoft’s Diagnostic Data Viewer page draws the same line for its own tool: seeing an event “doesn’t necessarily mean it has been uploaded yet”.
Is coding agent tracking new in Windows 11’s optional diagnostic data?
The bucket is not new; the named slots are. Microsoft’s optional diagnostic data reference already lists, under app usage, “Time of and count of app and component launches, duration of use, session GUID, and process ID”. By that wording, a machine sending optional data was already allowed to report when claude.exe started and for how long. The SubCohorts bitmap carries less than that.
What is new is the naming. The generic app-usage stream treats every executable alike. CliAgentSubCohorts gives Copilot, Claude Code, Codex and Antigravity fixed positions, which is the layout you would build to count which coding agents run on Windows machines, and it sits in the OS of a company that sells one of them. Microsoft has not said what the slots are for.
What does Microsoft’s required event list say about MissionControlAggregator?
It documents a second event from the same provider, and this one is required. The page Required diagnostic events and fields for Windows 11, versions 25H2 and 24H2 lists, under “Other events”, this entry (the grammar is Microsoft’s):
Microsoft.Windows.MissionControlAggregator.Segments.Cohorts
This events classifies users into Cohorts based on App usage - IsDeveloper
is the only cohort being reported.Its fields are CohortDetected, CohortEnabled, StartTime and Version. The 26H1 page has the same entry. The lists for 23H2 and 22H2, and for 21H2, do not mention MissionControlAggregator at all. Li’s build 26200 is 25H2, which the first page covers. His writeup does not mention this event, and the SubCohorts event he found appears on none of the required lists I read, which fits its "isRequired": false.
What the documentation does not tell you: which signals feed IsDeveloper (the agent slots, the tool categories, or something else), whether the event fires on a given machine, and when the entry was added. The 25H2/24H2 page carries an April 2026 date, but that is the page’s date, not the entry’s. It does tell you that the optional-data toggle does not cover this event.
How do you check whether MissionControlAggregator runs on your Windows 11 PC?
Three checks, cheapest first. I have not run these; they follow Li’s paths and Microsoft’s documented tools.
The state folder. From an elevated PowerShell:
Get-ChildItem 'C:\ProgramData\Microsoft\Diagnosis\AggregatorStorage' -ForceA MissionControlAggregator$ entry tells you the component has saved state on this machine at some point. It does not tell you which of the two events that state feeds, so it cannot confirm the per-agent feature is on. Absence is weaker still: the feature is switched on from Microsoft’s side and can arrive later.
Your diagnostic data level. Settings > Privacy & security > Diagnostics & feedback. Li’s clean Windows 11 Home install, set up in the US with the defaults, showed the toggle already on and labeled “Include Optional”. Microsoft’s configuration page says Required has been the default since Windows 10, version 1903, but that page applies to Pro, Enterprise and Education. Both can be true on different editions and setup paths. Look at your own machine instead of trusting either.
The Diagnostic Data Viewer. Under Settings > Privacy & security > Diagnostics & feedback > View diagnostic data, turn on Turn on the Diagnostic Data Viewer and install the viewer from the Microsoft Store. Use the machine normally for a day or two, since the report is roughly daily, then search for MissionControlAggregator and read the event name on each hit. Segments.SubCohorts is the per-agent bitmap Li found; check its CliAgentSubCohorts and DesktopAgentSubCohorts values. Segments.Cohorts is the required developer flag and says nothing about which agents ran. Microsoft calls the JSON shown “the exact details uploaded to Microsoft” and also says a listed event may still be waiting in the queue, so read a hit as queued or sent. An empty search does not prove the feature is off.
Without the Store, the PowerShell module does the same from an elevated session:
Install-Module -Name Microsoft.DiagnosticDataViewer
Enable-DiagnosticDataViewing
# a day or two later:
Get-DiagnosticData -StartTime (Get-Date).AddDays(-3) | Export-Csv "$env:TEMP\diag.csv"
Select-String -Path "$env:TEMP\diag.csv" -Pattern 'MissionControlAggregator'Data viewing can use up to 1 GB of disk. Microsoft recommends turning it off when you are done, which also clears the local history (Disable-DiagnosticDataViewing).
How do you turn off the Windows 11 coding agent report?
Turn off optional diagnostic data. Li observed the per-agent report queued with optional data allowed and tagged "isRequired": false; at the Required level, Windows is documented to send required events only. Nobody has published a run with the toggle off, so this is the documented behavior, not an observed one.
- Any edition: Settings > Privacy & security > Diagnostics & feedback, turn off Send optional diagnostic data (the same toggle Li’s Home install labeled “Include Optional”). If you are installing Windows, decline optional data in setup.
- Pro and up, by policy: Group Policy, Computer Configuration > Administrative Templates > Windows Components > Data Collection and Preview Builds > Allow diagnostic data: set it to Enabled and choose the Required level, value 1. Through MDM it is
System/AllowTelemetry. The same policy is stored as a REG_DWORD namedAllowTelemetryunder the key below; set it to 1.
HKEY_LOCAL_MACHINE\Software\Policies\Microsoft\Windows\DataCollectionThe policy values are 0 off, 1 Required and 3 Optional. Microsoft’s own registry example uses 0, but “Diagnostic data off” is available only on Enterprise, Education and Server editions. On Home and Pro the floor is Required.
Then, on the same Settings page, Delete diagnostic data removes what Microsoft holds for the device. It does not stop future collection, and it does not touch data tied to your Microsoft account, which lives in the account privacy dashboard.
What stays on after you turn off optional diagnostic data in Windows 11?
Everything required, and on Home and Pro you cannot go lower. Per Microsoft’s own list, that includes the Segments.Cohorts event with its IsDeveloper flag. With optional data off, “Codex CLI and a compiler ran in this window” should no longer go out. What can still go out is “this user is a developer”, if the event fires on your machine at all, which the documentation does not say. Required data also covers more than this one flag, from installed apps for compatibility checks to app state changes. The per-agent counters stay compiled in either way, and Microsoft decides from its side whether they run.
The only documented way below Required is value 0 on Enterprise, Education or Server. Blocking endpoints is outside anything Microsoft supports: its configuration page lists a settings endpoint it uses “to remotely block an event from being sent back to Microsoft”, and says not to block it.
Does Windows 11’s diagnostic data setting cover Claude Code, Codex or Cursor telemetry?
No. They are separate switches, and Microsoft says so: Windows diagnostic settings “only apply to components, features, and apps that are considered a part of the Windows operating system”, and third-party apps “may also collect and send diagnostic data using their own controls”. Windows learns that claude.exe ran. The agent’s own telemetry can say far more about what it did, and its controls live in the agent.
In Claude Code, telemetry off is DISABLE_TELEMETRY=1. Since 2.1.319, a session with telemetry off whose auto mode actions the server cannot review, typically one behind an LLM gateway or proxy, no longer falls back to Claude Code’s own classifier: those calls are asked about or denied. Check that before you set it on a machine that runs unattended.
The “disable all Windows telemetry” scripts going around on GitHub are not needed to opt out of the optional report. What they add is endpoint blocking, stopped services and hosts-file edits, which go after the required events the toggle leaves, by routes Microsoft does not support and whose side effects depend on what each script changes. So far the record is thin on both sides. The per-agent report is optional and one documented toggle should stop it; the developer flag is required and that toggle does not reach it on Home and Pro; Microsoft has explained neither. The agent’s own outbound data is a separate problem with separate controls, which the email-in-User-Agent piece works through for Claude Code.
01 / Discussion
No comment section here — all discussions happen on X.
02 / More articles
Claude Code "Don't ask again" and settings.local.json: how to find the passwords in your allow rules, and what to change
On the current Claude Code release the prompt turns most of a command into a wildcard before it saves the rule, and a password written in front of the program still survives. The session transcript keeps the whole command, and the cleanest fix is a command line that never holds the secret.
Claude Code and your email in a User-Agent: what the privacy mod checks, and what a hook and an allowlist add
Nobody built a feature that mails your address to websites. The model finds it in its context, meets a site that asks scripted clients for a contact, and helps. Every fix on offer filters that address on its way out, so rank the filters by what the model cannot argue past, and know which routes each one leaves open.
Prompt injection examples against coding agents: where the text got in, and what, if anything, stopped it
Nine public cases, sorted by the door the attacker's words came through. In most of them nothing held during the attack. The fixes that came after were all of one kind: a deterministic check re-aimed or a capability removed. A model's judgment that failed had nothing to re-aim.