streams ending mid-response with had_error=true across multiple repos (2.1.141.650)

Streams are dying mid-response, repeatedly. I send a message, the stream starts, a few hundred bytes come back, then it just ends with no completion. Happens across multiple repos in VS Code, not just one project, so it doesn't look workspace specific. Probably every app where I'm using the Claude Code VS Code extension.

The signal in the log is:
'
sdk_stream_ended_no_result {"had_error":true,"subagent_count":0,"message_count":167}
'

Followed by stall warnings as the watchdog notices nothing else is coming:

[WARN] [Stall] stream_idle_partial lastChunkAgeMs=15003 bytesTotal=691 idleDeadlineMs=300000
[WARN] Streaming stall detected: 71.9s gap between events (stall #1)

The stream begins fine ("Stream started - received first chunk") and then dies about 8 seconds in. Total bytes received is tiny, around 600 to 700.

Nothing surfaces in the UI. The session just goes idle. From the user side it looks like the assistant said nothing at all.

Environment:

• cc_version=2.1.141.650
• VS Code Claude Code extension
• Windows 11 Pro
• Model: Opus 4.7 (claude-opus-4-7[1m])
• Deep context conversations (message_count ~160+), but not sure

that's the trigger

• autocompact reports level=ok, effectiveWindow=980000, so it's not a

token ceiling

What I've checked:
It's not repo specific. Same behavior in multiple repos.
Reloading the extension doesn't reliably fix it well enough.
Some repos it's a worse issue than others but it's an issue across all of them at the moment.
Some chats are more affected than others.
Switching model has not been a consistent workaround.

Happy to attach a full log slice if you want one. The stall and the sdk_stream_ended_no_result event repro cleanly, just inconsistently on timing.