[BUG] Claude Desktop (Windows/MSIX): repeated silent process termination preceded by PerAppRuntimeBroker DCOM activation denial; first occurrence left package unlaunchable and destroyed Code tab history
Preflight Checklist
- [x] I have searched existing issues and this hasn't been reported yet
- [x] This is a single bug report (please file separate reports for different bugs)
- [x] I am using the latest version of Claude Code
What's Wrong?
<html>
<body>
<!--StartFragment--><html><head></head><body><h1>Claude Desktop (Windows/MSIX): repeated silent process termination preceded by PerAppRuntimeBroker DCOM activation denial; first occurrence left package unlaunchable and destroyed Code tab history</h1>
<h2>Summary</h2>
<p>On 2026-08-17, Claude Desktop terminated silently three times in roughly one hour. Each termination is preceded by 3–4 seconds by a DCOM event 10016 denying Local Activation on <strong>PerAppRuntimeBroker</strong> to a caller identified as running in the Claude application container.</p>
<p>The terminations produce <strong>no Windows Error Reporting entry and no Application Error event</strong> for <code>claude.exe</code>, so these are not ordinary faults — the process disappears without a crash record.</p>
<p>The first occurrence additionally left the MSIX package unlaunchable (Windows offered only "Repair / reinstall"). Repair failed, and the reinstall that followed permanently destroyed the Code tab conversation history index. The app crashed again twice on the fresh install, so the reinstall did not address the underlying cause.</p>
<p>The Claude Code CLI on the same machine is unaffected throughout.</p>
<h2>Environment</h2>
|
-- | --
OS | Windows (machine FSS-01), user Admin, PowerShell 7.6.3
Claude Desktop | Claude_1.30096.5.0_x64__pzs8sxrjxfjjc (MSIX)
Claude Code CLI | v2.1.181
Package originally installed | 2026-08-06 (clean; no deployment activity until the events below)
Working directory in use | C:\CMMS\dev\cmms
<h2>Timeline (2026-08-17, local time)</h2>
<p><strong>11:08:07</strong> — Claude VM Service starts normally.
<strong>11:08:43</strong> — App connects, signature verified. Session runs normally for ~1.5 hours.</p>
<h3>Occurrence 1</h3>
<p><strong>12:41:25</strong> — DCOM event 10016 (PerAppRuntimeBroker activation denied).
<strong>12:41:28.965</strong> — <code>cowork-service.log</code>: <code>[Server] Persistent RPC: connection ended: failed to read length: EOF</code>. App process gone. No WER, no Application Error.
<strong>12:41:38</strong> — Claude VM Service stops cleanly.</p>
<p>App would not relaunch. Windows returned: <em>"You'll need to go to advanced options for Claude and select Repair. If you're still having issues with it, reinstall this app or contact your administrator."</em> A restart did not help.</p>
<p><strong>12:43:58</strong> — Repair (<code>RegisterByPackageFullName</code> with <code>RepairAppRegistrationOption</code>). Log ends at <code>Trying to repair ACLs for \\?\C:\Program Files\WindowsApps\Claude_1.30096.5.0_x64__pzs8sxrjxfjjc</code> with no success entry; errors 867 / 763 / 651 and warnings 728 follow.
<strong>12:44:35</strong> — Uninstall.
<strong>12:49:51–12:49:56</strong> — Reinstall of the <em>same</em> version (1.30096.5.0), successful.</p>
<h3>Occurrence 2</h3>
<p><strong>13:30:06</strong> — DCOM event 10016, message quoted below.
<strong>13:30:10.926</strong> — Same <code>EOF</code> termination signature. App was reading <code>C:\CMMS\Rebuild\qab-handoff.md</code> and had just displayed a browser view.</p>
<p>App relaunched successfully afterwards; package remained intact.</p>
<h3>Occurrence 3</h3>
<p>Same pattern, shortly after Occurrence 2. <em>(Timestamp to be filled in.)</em></p>
<h2>The DCOM event</h2>
<pre><code>The application-specific permission settings do not grant Local Activation permission
for the COM Server application with CLSID
{2593F8B9-4EAF-457C-B68A-50F6B8EA6B54}
and APPID
{15C20B67-12E7-4BB6-92BB-7AFF07997402}
to the user FSS-01\Admin SID (S-1-5-21-...-1003) from address LocalHost (Using LRPC)
running in the application container Claude_1.30096.5.0_x64__pzs8sxrjxfjjc SID (Unavailable).
</code></pre>
<p>Both GUIDs resolve to <strong>PerAppRuntimeBroker</strong>.</p>
<p>Two details worth attention:</p>
<ol>
<li>The container SID is reported as <code>(Unavailable)</code>, suggesting the app container identity was not resolving at that moment.</li>
<li>Both terminations occurred while the app was crossing the sandbox boundary — reading a file outside package storage, and opening a browser view.</li>
</ol>
<p>Event 10016 fires ~32 times across 8/16–8/17 on this machine, so it is not rare in isolation. What is notable is that the instances immediately preceding <em>both</em> terminations are attributed to the Claude container, and there are no other System-log events in either window.</p>
<h2>What was ruled out</h2>
<ul>
<li><strong>Windows Defender / AV</strong> — <code>Get-MpThreatDetection</code> empty; no Defender operational events in the windows.</li>
<li><strong>Failed auto-update or deployment</strong> — the AppX deployment log contains <strong>no Claude entries at all</strong> between the 8/6 install and the 12:43 repair attempt. Same version before and after reinstall.</li>
<li><strong>Application crash</strong> — no WER report, no Application Error event, no faulting module for <code>claude.exe</code>. Silent termination, not a fault.</li>
<li><strong>VSS / shadow copy</strong> — the only VSS error (<code>0x8007045b</code>) is an artifact of the user-initiated reboot.</li>
<li><strong>Reinstall as a fix</strong> — occurrences 2 and 3 happened on a clean install.</li>
</ul>
<h2>Data loss (separate but consequential)</h2>
<p>The reinstall in occurrence 1 destroyed the Code tab Recents index. The app does not rebuild it by scanning <code>~/.claude/projects/</code>, so all prior Code conversations became unreachable from the UI even though every <code>.jsonl</code> transcript survived intact with correct timestamps.</p>
<p>There is no in-app recovery path. Sessions could only be reached via <code>claude --resume <session-id></code> from the CLI, which required knowing the session ID from the filesystem.</p>
<p>Two requests here:</p>
<ul>
<li>Code tab history should be as durable as regular Chat history, which survived the reinstall because it is account-tied.</li>
<li>Failing that, the Recents index should be reconstructible from the JSONL files already on disk.</li>
</ul>
<h2>Impact</h2>
<p>Claude Desktop is currently unusable for filesystem-heavy work on this machine. The Claude Code CLI is unaffected — it is not a packaged app, does not run in an app container, and never invokes the broker — which is consistent with the broker being implicated.</p>
<h2>Suggested investigation</h2>
<p>Whether the app handles a failed <code>PerAppRuntimeBroker</code> activation gracefully, or exits. A denied broker activation should surface as an error to the user, not terminate the process silently.</p>
<h2>Artifacts available on request</h2>
<ul>
<li><code>C:\ProgramData\Claude\Logs\cowork-service.log</code> (full)</li>
<li>AppXDeploymentServer/Operational export covering 8/6–8/17</li>
<li>System and Application log exports for both termination windows</li>
</ul>
<h2>Note</h2>
<p><code>cowork-service.log</code> also shows <code>[HCS] Warning: Failed to load vmcompute.dll: The specified module could not be found</code> on every startup (Hyper-V compute layer absent). Believed unrelated — it predates these events and appears on healthy startups — but noted for completeness.</p></body></html><!--EndFragment-->
</body>
</html>
What Should Happen?
All written in the description
Error Messages/Logs
NoneSteps to Reproduce
All written in the description
Claude Model
None
Is this a regression?
Yes, this worked in a previous version
Last Working Version
_No response_
Claude Code Version
v2.1.181
Platform
Anthropic API
Operating System
macOS
Terminal/Shell
Terminal.app (macOS)
Additional Information
_No response_