[BUG] Linux: "import cookies" in the in-app browser wedges the main webview, which is killed and never respawned (permanent black window, app still running)
Status Open
Reported on v2.1.280
Maintainer reply None cached
Activity 1 comment · opened Sep 25, 2026
Environment
- Claude Desktop on Linux (X11 / Cinnamon-Muffin), Electron app installed at
/usr/lib/claude-desktop,--ozone-platform=x11 [CCD] Initialized with version 2.1.280(from~/.config/Claude/logs/main.log)
What happened
Opened a GitHub pull request in the in-app browser and chose the import cookies option. The in-app browser wedged, then the whole window became an inert black surface while the app kept running headlessly.
Timeline from ~/.config/Claude/logs/:
15:11:30renderer--renderer-client-id=17created for the in-app browser;15:11:35–15:11:40it loads the github.com PR page (the only two lines ever written toin-app-browser.logaregithub.githubassets.compreload warnings)- chose import cookies
15:12:08main.log:Main webview is unresponsive, will kill and reload- no new renderer process is ever created. Both pre-existing renderers survive that line — client-id 6 (main window, started
15:06:17) and client-id 17 (in-app browser,15:11:30) — and the main window paints nothing from then on claude.ai-web.logstops at15:06:20(app start); the web layer emits nothing during or after the freeze- nothing anywhere in
main.logmentions cookies or the import itself
Symptoms
- The main window renders a uniform black surface, no interaction possible. Verified objectively with
xwd: the 8 MB capture gzips to 21 KB (i.e. uniform), versus 168 KB for the same window after a healthy restart. - Everything except the UI keeps working: main process, GPU process, utility processes, MCP servers and the local Claude Code sessions.
main.logkept logging PR polling,[process-memory]intervals, and at15:17:07evenEmitted tool permission request … for AskUserQuestion— a session asking a question that can never be answered, because there is no UI to answer it in. - Desktop notifications stop arriving; they surface in the browser's claude.ai tab instead.
- Memory at the time:
tree_rss_sum≈7150MB,electron(10)=1434MB,children(74)=5718MB— no OOM (sys_free=17.5GB/31.8GB).
Recovery attempts that did not work (app left running throughout)
- Window resize nudge to force relayout/repaint — still uniform.
kill -TERMon the in-app browser renderer (client-id 17) — still uniform.kill -TERMon the main webview renderer (client-id 6) — surface changed from black to blank, and no renderer was respawned. Process inventory dropped to1 gpu-process, 6 utility, 3 zygote, 0 renderers, and stayed there.- Second-instance handoff (
/usr/bin/claude-desktoplaunched again while the first instance heldSingletonLock).main.log:second-instance: suppressing duplicate argv,Not main instance, returning early from app ready, thenmainView bounds were stale on focus — resyncing (wayland resize race?)— note this session is X11, not Wayland. Still no renderer.
Only a full restart recovered it. On restart: [CCD-SessionRecovery] Scan complete: 54 recoverable and the UI came back normally.
Expected
- Importing cookies should not wedge the in-app browser webview.
- When the app logs
Main webview is unresponsive, will kill and reload, the reload should actually produce a new renderer process. Today the app reaches a state with zero renderers and never self-heals, and there is no visible reload affordance on the black window — the only exit is killing the app, which takes every in-flight local session with it, including ones blocked on a permission prompt.
Possibly related
- #87764 — Desktop Browser pane has no way to use existing browser login state
- #95088 — Chrome cookie import fails on macOS (Safe Storage key in the wrong keychain)
This issue has 1 comment on GitHub. Read the full discussion on GitHub ↗