@jsonstudio/llms 0.6.97 → 0.6.104

package/dist/conversion/hub/pipeline/stages/resp_outbound/resp_outbound_stage1_client_remap/index.js

@@ -62,4 +62,16 @@ function mergeOriginalResponsesPayload(payload, adapterContext) {
     if (rawStatus === 'requires_action') {
         payload.status = 'requires_action';
     }
+    // 如果桥接后的 payload 没有 usage，而原始 Responses 载荷带有 usage，则回填原始 usage，
+    // 确保 token usage 不在工具/桥接路径中丢失。
+    const payloadUsage = payload.usage;
+    const rawUsage = raw.usage;
+    if ((payloadUsage == null || typeof payloadUsage !== 'object') && rawUsage && typeof rawUsage === 'object') {
+        try {
+            payload.usage = JSON.parse(JSON.stringify(rawUsage));
+        }
+        catch {
+            payload.usage = rawUsage;
+        }
+    }
 }

package/dist/router/virtual-router/engine.d.ts

@@ -35,4 +35,5 @@ export declare class VirtualRouterEngine {
     private buildRouteCandidates;
     private sortByPriority;
     private routeWeight;
+    private buildHitReason;
 }

package/dist/router/virtual-router/engine.js

@@ -35,7 +35,8 @@ export class VirtualRouterEngine {
         const target = this.providerRegistry.buildTarget(selection.providerKey);
         this.healthManager.recordSuccess(selection.providerKey);
         this.incrementRouteStat(selection.routeUsed, selection.providerKey);
-        this.debug?.log?.('[virtual-router-hit]', selection.routeUsed, selection.providerKey, target.modelId || '');
+        const hitReason = this.buildHitReason(selection.routeUsed, classification, features);
+        this.debug?.log?.('[virtual-router-hit]', selection.routeUsed, selection.providerKey, target.modelId || '', hitReason ? `reason=${hitReason}` : '');
         const didFallback = selection.routeUsed !== routeName || classification.fallback;
         return {
             target,
@@ -179,22 +180,23 @@ export class VirtualRouterEngine {
         const code = event.code?.toUpperCase() ?? 'ERR_UNKNOWN';
         const stage = event.stage?.toLowerCase() ?? 'unknown';
         const recoverable = event.recoverable === true;
-        // 默认策略：只有显式可恢复的错误才视为非致命；其余一律按致命处理
+        // 默认策略：只有显式可恢复的错误才视为非致命；其余一律按致命处理。
+        // 注意：provider 层已经对 429 做了「连续 4 次升级为不可恢复」的判断，这里不再把所有 429 强行当作可恢复。
         let fatal = !recoverable;
         let reason = this.deriveReason(code, stage, statusCode);
         let cooldownOverrideMs;
-        // 400 / 429 作为明确可恢复池：走限流通道，不做长期拉黑
-        if (statusCode === 429 || code.includes('429') || statusCode === 400 || code.includes('400')) {
-            fatal = false;
-            cooldownOverrideMs = Math.max(30_000, this.providerHealthConfig().cooldownMs);
-            reason = 'rate_limit';
-            // 401 / 402 / 500 / 524 以及所有未被标记为可恢复的错误一律视为不可恢复
-        }
-        else if (statusCode === 401 || statusCode === 402 || statusCode === 403 || code.includes('AUTH')) {
+        // 401 / 402 / 500 / 524 以及所有未被标记为可恢复的错误一律视为不可恢复
+        if (statusCode === 401 || statusCode === 402 || statusCode === 403 || code.includes('AUTH')) {
             fatal = true;
             cooldownOverrideMs = Math.max(10 * 60_000, this.providerHealthConfig().fatalCooldownMs ?? 10 * 60_000);
             reason = 'auth';
         }
+        else if (statusCode === 429 && !recoverable) {
+            // 连续 429 已在 provider 层被升级为不可恢复：这里按致命限流处理（长冷却，等同熔断）
+            fatal = true;
+            cooldownOverrideMs = Math.max(10 * 60_000, this.providerHealthConfig().fatalCooldownMs ?? 10 * 60_000);
+            reason = 'rate_limit';
+        }
         else if (statusCode && statusCode >= 500) {
             fatal = true;
             cooldownOverrideMs = Math.max(5 * 60_000, this.providerHealthConfig().fatalCooldownMs ?? 5 * 60_000);
@@ -213,7 +215,8 @@ export class VirtualRouterEngine {
             statusCode,
             errorCode: code,
             retryable: recoverable,
-            affectsHealth: true,
+            // 是否影响健康由 provider 层决定；这里仅在 event.affectsHealth !== false 时才计入健康状态
+            affectsHealth: event.affectsHealth !== false,
             cooldownOverrideMs,
             metadata: {
                 ...event.runtime,
@@ -274,4 +277,25 @@ export class VirtualRouterEngine {
         const idx = ROUTE_PRIORITY.indexOf(routeName);
         return idx >= 0 ? idx : ROUTE_PRIORITY.length;
     }
+    buildHitReason(routeUsed, classification, features) {
+        const reasoning = classification.reasoning || '';
+        const primary = reasoning.split('|')[0] || '';
+        const lastToolName = features.lastAssistantToolName;
+        if (routeUsed === 'tools') {
+            if (lastToolName) {
+                return primary ? `${primary}(${lastToolName})` : `tools(${lastToolName})`;
+            }
+            return primary || 'tools';
+        }
+        if (routeUsed === 'thinking') {
+            return primary || 'thinking';
+        }
+        if (routeUsed === DEFAULT_ROUTE && classification.fallback) {
+            return primary || 'fallback:default';
+        }
+        if (primary) {
+            return primary;
+        }
+        return routeUsed ? `route:${routeUsed}` : 'route:unknown';
+    }
 }

package/dist/router/virtual-router/types.d.ts

@@ -210,6 +210,7 @@ export interface ProviderErrorEvent {
     stage: string;
     status?: number;
     recoverable?: boolean;
+    affectsHealth?: boolean;
     runtime: ProviderErrorRuntimeMetadata;
     timestamp: number;
     details?: Record<string, unknown>;

package/dist/test-output/virtual-router/results.json

@@ -0,0 +1 @@
+[]

package/dist/test-output/virtual-router/summary.json

@@ -0,0 +1,12 @@
+{
+  "samplesRoot": "/Users/fanzhang/.routecodex/codex-samples",
+  "configPath": "/Users/fanzhang/Documents/github/sharedmodule/llmswitch-core/test/virtual-router/virtual-router.config.json",
+  "stats": {
+    "totalSamples": 0,
+    "processed": 0,
+    "routes": {},
+    "providers": {},
+    "errors": [],
+    "scenarios": {}
+  }
+}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@jsonstudio/llms",
-  "version": "0.6.097",
+  "version": "0.6.104",
   "type": "module",
   "main": "dist/index.js",
   "module": "dist/index.js",