fix(agents): avoid overload classification for live model switches

CHANGELOG.md

@@ -21,6 +21,7 @@ Docs: https://docs.openclaw.ai
 ### Fixes
 
 - CLI/doctor: run bundled plugin runtime-dependency repairs through the async npm installer with spinner/line progress and heartbeat updates, so long `openclaw doctor --fix` installs no longer look hung in TTY or piped output. Fixes #72775. Thanks @dfpalhano.
+- Agents/fallback: classify internal live-session model switch conflicts as unknown fallback failures instead of provider overloads, preventing local vLLM endpoints from receiving misleading overloaded cooldowns. Refs #63229. Thanks @clawdia-lobster.
 - Control UI: keep session-specific assistant identity loads authoritative after WebSocket connect, so non-main agent chat sessions do not show the main agent name in the header after bootstrap refreshes. Fixes #72776. Thanks @rockytian-top.
 - Agents/Qwen: preserve exact custom `modelstudio` provider configs with foreign `api` owners so explicit OpenAI-compatible Model Studio endpoints no longer get normalized into the bundled Qwen plugin path. Fixes #64483. Thanks @FiredMosquito831.
 - Media-understanding/audio: migrate deprecated `{input}` placeholders in legacy `audio.transcription.command` configs to `{{MediaPath}}`, so custom audio transcribers no longer receive the literal placeholder after doctor repair. Fixes #72760. Thanks @krisfanue3-hash.

src/agents/model-fallback.test.ts

@@ -686,6 +686,7 @@ describe("runWithModelFallback", () => {
     // Should NOT be a LiveSessionModelSwitchError — the outer retry loop must
     // not restart with the conflicting model.
     expect(err).not.toBeInstanceOf(LiveSessionModelSwitchError);
+    expect((err as { reason?: string }).reason).toBe("unknown");
     expect(run).toHaveBeenCalledTimes(1);
   });
 
@@ -774,6 +775,7 @@ describe("runWithModelFallback", () => {
     expect(result.result).toBe("ok");
     expect(result.provider).toBe("anthropic");
     expect(result.model).toBe("claude-haiku-3-5");
+    expect(result.attempts[0]?.reason).toBe("unknown");
     expect(run.mock.calls).toEqual([
       ["openai", "gpt-4.1-mini"],
       ["anthropic", "claude-haiku-3-5"],

src/agents/model-fallback.ts

@@ -936,7 +936,8 @@ export async function runWithModelFallback<T>(params: {
       // LiveSessionModelSwitchError during fallback may point at a later
       // candidate that is already the active live-session selection.  Jump
       // there directly.  Stale same/earlier targets remain a known failover
-      // so the outer runner cannot loop on the conflicting model.
+      // so the outer runner cannot loop on the conflicting model, but they
+      // are not provider overloads.
       if (err instanceof LiveSessionModelSwitchError) {
         const liveSwitchTargetIndex = findLiveSessionModelSwitchRedirectIndex({
           error: err,
@@ -950,7 +951,7 @@ export async function runWithModelFallback<T>(params: {
 
         const switchMsg = err.message;
         const switchNormalized = new FailoverError(switchMsg, {
-          reason: "overloaded",
+          reason: "unknown",
           provider: candidate.provider,
           model: candidate.model,
         });