fix(auth): lower transient error cooldown default to 10s

config.example.yaml

@@ -167,7 +167,7 @@ disable-cooling: false
 save-cooldown-status: false
 
 # Cooldown duration in seconds for transient upstream errors (408/500/502/503/504).
-# Set to 0 to keep the legacy 60-second cooldown; set to -1 to disable transient error cooldowns.
+# Set to 0 to keep the legacy 10-second cooldown; set to -1 to disable transient error cooldowns.
 transient-error-cooldown-seconds: 0
 
 # When true, globally disable Claude request cloaking (the Claude Code CLI disguise and

sdk/cliproxy/auth/conductor_overrides_test.go

@@ -946,8 +946,8 @@ func TestManager_MarkResult_TransientErrorCooldownDefault(t *testing.T) {
 		t.Fatal("expected transient error cooldown to keep the legacy default")
 	}
 	diff := time.Until(state.NextRetryAfter)
-	if diff < 55*time.Second || diff > 65*time.Second {
-		t.Fatalf("expected transient error cooldown to be ~60 seconds, got %v", diff)
+	if diff < 5*time.Second || diff > 15*time.Second {
+		t.Fatalf("expected transient error cooldown to be ~10 seconds, got %v", diff)
 	}
 }
 

sdk/cliproxy/auth/conductor_refresh.go

@@ -32,7 +32,12 @@ const (
 	refreshIneffectiveBackoff = 30 * time.Second
 	quotaBackoffBase          = time.Second
 	quotaBackoffMax           = 30 * time.Minute
-	transientErrorCooldown    = time.Minute
+	// transientErrorCooldown is the default for 408/500/502/503/504 transient
+	// errors and the fallback enforced cooldown for request-scoped
+	// stop-and-cooldown / continue-and-cooldown rules. The latter use this
+	// constant directly and do not consult the transient-error-cooldown-seconds
+	// knob.
+	transientErrorCooldown = 10 * time.Second
 )
 
 // StartAutoRefresh launches a background loop that evaluates auth freshness