- 修复了 qwen 图像测试报 unknown image provider：PingImage 现在按 drivers.image 分…

.specify/memory/api-naming.md

@@ -43,8 +43,7 @@
 ## 3. 鉴权与租户透传
 
 - **所有 `/api/v1/admin` 与 `/api/internal` 必须鉴权**
-- 租户信息必须通过 token 或 header 解析：
-  - `tenant_uuid` / `X-PowerX-Tenant`
+- 租户信息必须通过 token（JWT claims）或 `tenant_uuid` 字段解析，不接受遗留租户头注入。
 - 内部接口也需 tenant 校验，禁止跨租户调用
 
 ---

.specify/memory/dev_crud_grpc_guides.md

@@ -66,7 +66,7 @@ internal/transport/grpc/
 
 ### 5.1 绑定/校验
 
-* **tenant 提取**：优先读 `RequestContext.tenant_id`，其次从 Metadata 头 `x-powerx-tenant-id|tenant-id|x-tenant-id` 兜底。
+* **tenant 提取**：优先读 `RequestContext.tenant_id`，其次仅从 Metadata `tenant-id` 兜底（不接受任何 `x-powerx-*` 遗留租户头）。
   你的 `tenantIDFrom()` 已经实现这一落地逻辑。
 * **分页映射**：`PageRequest(offset,page_size)` → `(page,size)` 的换算统一用工具函数（如 `pageFrom()`）。
 * **错误回包**：Meta 中返回 `code/message/request_id`，与你的 `okMeta/badMeta` 一致（见 `member_handler.go`/`team_handler.go` 调用）。
@@ -140,4 +140,3 @@ internal/transport/grpc/
 * [ ] **与 HTTP 等价**：同一用例在 HTTP 与 gRPC 的语义、错误、分页完全可对照。
 
 ---
-

backend/config/agents/providers.d/anthropic.yaml

@@ -6,7 +6,8 @@ drivers:
   llm: anthropic
   embedding: anthropic
   image: anthropic
-  audio: anthropic
+  audio_tts: anthropic
+  audio_asr: anthropic
   video: anthropic
   rerank: anthropic
 
@@ -49,9 +50,11 @@ modalities:
     models: []   # 有 Embeddings 能力，但当前以 Messages API 为主；如需可后续补充
   image:
     models: []
-  audio:
+  audio_tts:
+    models: []
+  audio_asr:
     models: []
   video:
     models: []
   rerank:
-    models: []
+    models: []

backend/config/agents/providers.d/baidu_wenxin.yaml

@@ -6,7 +6,8 @@ drivers:
   llm: baidu
   embedding: baidu
   image: baidu
-  audio: baidu
+  audio_tts: baidu
+  audio_asr: baidu
   video: baidu
   rerank: baidu
 
@@ -79,7 +80,9 @@ modalities:
         # 官方未公开统一维度，这里不写 dims，避免类型不符
   image:
     models: []
-  audio:
+  audio_tts:
+    models: []
+  audio_asr:
     models: []
   video:
     models: []

backend/config/agents/providers.d/comfyui.yaml

@@ -27,3 +27,7 @@ modalities:
   video:
     note: "ComfyUI Video（UI 目录占位）。"
     models: []
+  audio_tts:
+    models: []
+  audio_asr:
+    models: []

backend/config/agents/providers.d/coze.yaml

@@ -6,7 +6,8 @@ drivers:
   llm: coze
   embedding: coze
   image: coze
-  audio: coze
+  audio_tts: coze
+  audio_asr: coze
   video: coze
   rerank: coze
 
@@ -35,9 +36,11 @@ modalities:
     models: []   # 如你的 driver 支持 Coze 向量服务，可在此处补充
   image:
     models: []
-  audio:
+  audio_tts:
+    models: []
+  audio_asr:
     models: []
   video:
     models: []
   rerank:
-    models: []
+    models: []

backend/config/agents/providers.d/deepseek.yaml

@@ -6,7 +6,8 @@ drivers:
   llm: deepseek
   embedding: deepseek
   image: deepseek
-  audio: deepseek
+  audio_tts: deepseek
+  audio_asr: deepseek
   video: deepseek
   rerank: deepseek
 
@@ -31,9 +32,11 @@ modalities:
     models: []   # 暂未公开官方 Embedding API
   image:
     models: []
-  audio:
+  audio_tts:
+    models: []
+  audio_asr:
     models: []
   video:
     models: []
   rerank:
-    models: []
+    models: []

backend/config/agents/providers.d/google.yaml

@@ -20,3 +20,7 @@ modalities:
       - id: gemini-3-pro-image-preview
         label: "Gemini 3 Pro Image Preview (Nano Banana Pro)"
         tags: ["image", "high-quality"]
+  audio_tts:
+    models: []
+  audio_asr:
+    models: []

backend/config/agents/providers.d/hash.yaml

@@ -17,3 +17,7 @@ modalities:
         tags: ["local","dev","non-semantic"]
         defaults:
           dimensions: 1536
+  audio_tts:
+    models: []
+  audio_asr:
+    models: []

backend/config/agents/providers.d/huggingface.yaml

@@ -6,7 +6,8 @@ drivers:
   llm: huggingface
   embedding: huggingface
   image: huggingface
-  audio: huggingface
+  audio_tts: huggingface
+  audio_asr: huggingface
   video: huggingface
   rerank: huggingface
 
@@ -81,14 +82,16 @@ modalities:
       - id: stabilityai/stable-diffusion-2
         label: "Stable Diffusion 2 (text-to-image)"
         tags: ["img-gen","open"]
-  audio:
+  audio_tts:
     models:
-      - id: openai/whisper-large-v3
-        label: "Whisper Large v3 (ASR)"
-        tags: ["asr","open"]
       - id: coqui/XTTS-v2
         label: "Coqui XTTS v2 (TTS)"
         tags: ["tts","multilingual"]
+  audio_asr:
+    models:
+      - id: openai/whisper-large-v3
+        label: "Whisper Large v3 (ASR)"
+        tags: ["asr","open"]
   video:
     models: []
   rerank:

backend/config/agents/providers.d/hunyuan.yaml

@@ -126,3 +126,9 @@ modalities:
       - id: hunyuan-video
         label: "混元 视频生成 (placeholder)"
         tags: ["video","tencent","placeholder"]
+
+  audio_tts:
+    models: []
+
+  audio_asr:
+    models: []

backend/config/agents/providers.d/moonshot.yaml

@@ -6,7 +6,8 @@ drivers:
   llm: moonshot
   embedding: moonshot
   image: moonshot
-  audio: moonshot
+  audio_tts: moonshot
+  audio_asr: moonshot
   video: moonshot
   rerank: moonshot
 
@@ -39,9 +40,11 @@ modalities:
     models: []   # 官方暂未公开独立 Embedding API
   image:
     models: []
-  audio:
+  audio_tts:
+    models: []
+  audio_asr:
     models: []
   video:
     models: []
   rerank:
-    models: []
+    models: []

backend/config/agents/providers.d/openai.yaml

@@ -6,7 +6,8 @@ drivers:
   llm: openai
   embedding: openai
   image: openai
-  audio: openai
+  audio_tts: openai
+  audio_asr: openai
   video: openai
   rerank: openai
 
@@ -97,11 +98,13 @@ modalities:
         label: "DALL·E 2"
         tags: ["img-gen","legacy"]
 
-  audio:
+  audio_tts:
     models:
       - id: gpt-4o-mini-tts
         label: "GPT-4o mini TTS"
         tags: ["audio","tts","expressive"]
+  audio_asr:
+    models:
       - id: gpt-4o-transcribe
         label: "GPT-4o Transcribe"
         tags: ["audio","asr","multilingual"]

backend/config/agents/providers.d/openai_compatible.yaml

@@ -23,3 +23,7 @@ modalities:
       - id: "*"
         label: "Any (OpenAI-compatible)"
         tags: ["local","gateway"]
+  audio_tts:
+    models: []
+  audio_asr:
+    models: []

backend/config/agents/providers.d/qwen-cn.yaml

@@ -0,0 +1,51 @@
+id: qwen-cn
+name: Qwen（官方国内）
+aliases: ["qwen-mainland", "tongyi-cn", "dashscope-cn", "alibaba-cn"]
+
+drivers:
+  image: qwen
+  audio_tts: qwen
+
+auth:
+  scheme: bearer
+  fields: ["api_key", "base_url"]
+  defaults:
+    base_url: "https://dashscope.aliyuncs.com/compatible-mode/v1"
+
+modalities:
+  image:
+    models:
+      - id: qwen-vl-max-latest
+        label: "Qwen VL Max Latest"
+      - id: qwen-vl-plus-latest
+        label: "Qwen VL Plus Latest"
+      - id: qwen-vl-max
+        label: "Qwen VL Max"
+      - id: qwen-vl-plus
+        label: "Qwen VL Plus"
+      - id: qwen3-vl-plus
+        label: "Qwen3 VL Plus"
+      - id: qwen3-vl-flash
+        label: "Qwen3 VL Flash"
+      - id: qwen2.5-vl-72b-instruct
+        label: "Qwen 2.5 VL 72B Instruct"
+      - id: "*"
+        label: "Qwen VLM (Any)"
+  audio_tts:
+    models:
+      - id: qwen-tts-latest
+        label: "Qwen TTS Latest"
+      - id: qwen-tts-1
+        label: "Qwen TTS 1"
+      - id: qwen3-tts-flash
+        label: "Qwen3 TTS Flash"
+      - id: qwen3-tts-instruct-flash
+        label: "Qwen3 TTS Instruct Flash"
+      - id: qwen3-tts-vd
+        label: "Qwen3 TTS VD"
+      - id: qwen3-tts-vc
+        label: "Qwen3 TTS VC"
+      - id: "*"
+        label: "Qwen TTS (Any)"
+  audio_asr:
+    models: []

backend/config/agents/providers.d/qwen.yaml

@@ -0,0 +1,51 @@
+id: qwen-intl
+name: Qwen（官方国际）
+aliases: ["qwen", "tongyi", "dashscope", "alibaba"]
+
+drivers:
+  image: qwen
+  audio_tts: qwen
+
+auth:
+  scheme: bearer
+  fields: ["api_key", "base_url"]
+  defaults:
+    base_url: "https://dashscope-intl.aliyuncs.com/compatible-mode/v1"
+
+modalities:
+  image:
+    models:
+      - id: qwen-vl-max-latest
+        label: "Qwen VL Max Latest"
+      - id: qwen-vl-plus-latest
+        label: "Qwen VL Plus Latest"
+      - id: qwen-vl-max
+        label: "Qwen VL Max"
+      - id: qwen-vl-plus
+        label: "Qwen VL Plus"
+      - id: qwen3-vl-plus
+        label: "Qwen3 VL Plus"
+      - id: qwen3-vl-flash
+        label: "Qwen3 VL Flash"
+      - id: qwen2.5-vl-72b-instruct
+        label: "Qwen 2.5 VL 72B Instruct"
+      - id: "*"
+        label: "Qwen VLM (Any)"
+  audio_tts:
+    models:
+      - id: qwen-tts-latest
+        label: "Qwen TTS Latest"
+      - id: qwen-tts-1
+        label: "Qwen TTS 1"
+      - id: qwen3-tts-flash
+        label: "Qwen3 TTS Flash"
+      - id: qwen3-tts-instruct-flash
+        label: "Qwen3 TTS Instruct Flash"
+      - id: qwen3-tts-vd
+        label: "Qwen3 TTS VD"
+      - id: qwen3-tts-vc
+        label: "Qwen3 TTS VC"
+      - id: "*"
+        label: "Qwen TTS (Any)"
+  audio_asr:
+    models: []

backend/config/agents/providers.d/sentence-transformers.yaml

@@ -6,7 +6,8 @@ drivers:
   llm: sentence_transformers
   embedding: sentence_transformers
   image: sentence_transformers
-  audio: sentence_transformers
+  audio_tts: sentence_transformers
+  audio_asr: sentence_transformers
   video: sentence_transformers
   rerank: sentence_transformers
 
@@ -39,7 +40,9 @@ modalities:
         limits: { dims: 384 }
   image:
     models: []
-  audio:
+  audio_tts:
+    models: []
+  audio_asr:
     models: []
   video:
     models: []
@@ -50,4 +53,4 @@ modalities:
         tags: ["rerank","fast"]
       - id: cross-encoder/ms-marco-MiniLM-L-12-v2
         label: "MS MARCO MiniLM-L-12 (Cross-Encoder)"
-        tags: ["rerank","accurate"]
+        tags: ["rerank","accurate"]