You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
[Bug]: Claude runtime stream failed when closing T3Code #5323
Same failure on Linux, triggered by the nightly auto-update relaunch rather than a manual quit. Adding the event sequence since the original report has no logs.
Environment: CachyOS (Linux 7.2.5), T3 Code desktop 0.0.41-nightly.20260913.1658 auto-updating to 0.0.41-nightly.20260914.1700, 2.1.270 (Claude Code), Fable 5.1 thread, full-access runtime.
Sequence (UTC, from the thread's provider event log and the system journal):
10:39:53.899 turn.completed (end_turn) # last foreground turn, session idle from here
10:44:40.648 Claude CLI SessionEnd hook fires # subprocess exiting; T3 has not issued stopSession
10:44:40.850 runtime.error "Claude runtime stream failed." failureTags=["ProviderAdapterProcessError"]
10:44:40.850 claude/result status=failed # synthesized, no active turn
10:44:40.851 session.exited reason="Session stopped" exitKind="graceful"
10:44:42 new T3 server process starts (updated AppImage)
10:50:34 next user message: session.started with resume, turn runs normally
The updater's quit path SIGTERMs the Claude subprocess before the adapter's own stop runs, so the SDK stream rejects and the adapter reports it as a stream failure with no turn in flight. The thread is fine afterwards; the only effect is the error toast on reopen.
Activity
Nelglor commentedon Sep 14, 2026
Note
🤖 Claude Fable 5.1 responding on behalf of Nick
Same failure on Linux, triggered by the nightly auto-update relaunch rather than a manual quit. Adding the event sequence since the original report has no logs.
Environment: CachyOS (Linux 7.2.5), T3 Code desktop 0.0.41-nightly.20260913.1658 auto-updating to 0.0.41-nightly.20260914.1700, 2.1.270 (Claude Code), Fable 5.1 thread, full-access runtime.
Sequence (UTC, from the thread's provider event log and the system journal):
The updater's quit path SIGTERMs the Claude subprocess before the adapter's own stop runs, so the SDK stream rejects and the adapter reports it as a stream failure with no turn in flight. The thread is fine afterwards; the only effect is the error toast on reopen.
Two details that may help triage:
runtime.errorpayload keeps onlyfailureCountandfailureTags; the SDK's underlying process error text is dropped, so the "terminated by signal" message that fix(server): treat signal-killed Claude subprocess as interrupted #5142 matches on never reaches the logs. [Bug]: Claude session exits when a background Bash task stops after its turn completed #9698 notes the same loss for a different trigger and is probably the same mapping.exitKind: "graceful"immediately after the error, which is a cheap signal to suppress the toast when the app is shutting down.#5142 looks like the right fix and applies cleanly to this path.