Mid-stream api_error/overloaded_error SSE events (e.g. a local provider such
as LM Studio rejecting a malformed tool_call) arrive inside the 200 SSE body,
bypassing withRetry — which only guards stream creation — so they were
surfaced as a terminal API error that ended the turn with no retry.
Add withStreamRetry around queryModel: when an attempt produced zero output
(no tool_use block completed, so query.ts never started a tool — safe w.r.t.
the double-tool-execution hazard, inc-4258), re-establish the stream and retry
up to CLAUDE_STREAM_TRANSIENT_RETRY_MAX times (default 2) before surfacing the
error. Covers both the streaming and non-streaming query wrappers; adds unit
tests for the retry loop and the transient-error classifier.
Desktop official OAuth sessions inject a token into a long-lived CLI process to avoid OS credential prompts. After Windows sleep/wake that env token can become stale, so the server now pushes a refreshed token before each resumed user turn and the CLI clears its OAuth cache when the env token changes.
The retry path also disables keep-alive after ECONNRESET/EPIPE by default, matching the stale pooled-socket failure mode reported on #651.
Constraint: Real Win11 sleep/wake cannot be physically reproduced in this macOS worktree
Rejected: Restart the CLI before every user turn | would add latency and discard useful process continuity
Confidence: medium
Scope-risk: moderate
Directive: Keep official OAuth env-token updates paired with CLI-side OAuth cache invalidation
Tested: bun test src/server/__tests__/conversation-service.test.ts --timeout 30000
Tested: bun test src/cli/__tests__/structuredIO.test.ts --timeout 30000
Tested: bun test src/services/api/withRetry.test.ts --timeout 30000
Tested: bun run check:server
Tested: git diff --check
Not-tested: Real Win11 sleep/wake with Claude official OAuth
Not-tested: External ccswitch reqwest pool reset path when the reset is hidden behind the proxy
Related: #658
Related: #651
