chore(align): sync Git with running container as authoritative source

Standards alignment (STANDARDS_AUDIT_2026-07-24): Phase 1 标准化核心 +
Phase 2 质量层 landed in the running container and captured in Git so the
three sources (container := image := Git) agree.

Backend (Phase 1 标准化核心):
- mcp_client.py: replaced hand-rolled JSON-RPC with official `mcp` SDK
  (ClientSession, streamable_http/sse transports, protocol auto-negotiation)
- core/otel.py + core/observability.py: standard OpenTelemetry SDK bridge
  (OTLP gRPC), wired to Grafana Alloy (qihu-aiflow-alloy:4317) which forwards
  metrics to Prometheus; emits aiflow_span_duration_ms / aiflow_span_count
- deployments: Grafana Alloy container + Prometheus remote-write receiver

Backend (Phase 2 质量层):
- core/rag_pipeline.py: standardized RAG (recursive overlap chunking +
  rerank hook) reusing pgvector_search
- core/llm_client.py: SDK-ified LLM + embeddings (aembed/embed) collapsing
  hand-rolled httpx calls; trust_env=False to bypass container proxy

Migrations:
- 018_normalize_marketapp_scene.sql: fixed capability allowlist
  (prevents capability wipe on restart)

Alignment performed 2026-07-24 12:18:51. No deletions; secrets/venv/node_modules/
caches were not touched (tracked-source-only copy).

backend/python/app/core/llm_client.py

@@ -0,0 +1,181 @@
+"""Shared, standards-based LLM client for web chat + IM bot.
+
+平台红线(07-24)：模型层以成熟/开源标准为设计基准。本模块使用官方 OpenAI SDK
+(`openai.AsyncOpenAI`)——它是 OpenAI 兼容端点的事实标准客户端。平台本身已在讲
+OpenAI 协议(`/v1/chat/completions`)，因此零协议改造成本，并天然获得：
+  * 原生 SSE 流式(`stream=True`)；
+  * 结构化的 tool_calls / usage 解析；
+  * 统一的代理(trust_env=False + 显式 proxy)与超时处理。
+
+本模块取代 chat_service.py 与 im_llm.py 中两套手写 httpx 循环，使 web chat 与
+IM bot 共用**同一模型层**、同一套 failover/代理/超时语义——这是「各端能力统一对齐」
+红线在模型层的落地。
+
+API:
+  achat_complete(api_url, api_key, request_body, *, proxy=None, timeout=120, stream=False)
+    - stream=False -> dict，形状对齐上游 `data` 响应：
+        {"choices":[{"message":{"content","tool_calls","reasoning_content"}}], "usage":{...}}
+    - stream=True  -> async generator，逐块产出 ("content"|"reasoning", text) 增量
+"""
+from __future__ import annotations
+
+import httpx
+from openai import AsyncOpenAI, OpenAI
+
+
+def _base_url(api_url: str) -> str:
+    """从完整 /v1/chat/completions 或 /v1/embeddings 地址推导 OpenAI SDK 所需的 base_url。"""
+    u = (api_url or "").rstrip()
+    for suf in ("/v1/chat/completions", "/chat/completions", "/v1/embeddings", "/embeddings"):
+        if u.endswith(suf):
+            return u[: -len(suf)]
+    return u
+
+
+def build_openai_client(api_url, api_key, *, proxy=None, timeout=120) -> AsyncOpenAI:
+    base = _base_url(api_url)
+    # 始终 trust_env=False：避免容器 HTTP_PROXY 对国内端点(siliconflow 等)造成 502；
+    # 代理经 proxy 显式传入（AGNES 等），未传则直连。
+    http_client = httpx.AsyncClient(proxy=proxy, trust_env=False)
+    return AsyncOpenAI(
+        base_url=base,
+        api_key=api_key or "EMPTY",
+        timeout=timeout,
+        http_client=http_client,
+    )
+
+
+def _create_kwargs(request_body: dict) -> dict:
+    kw = {
+        "model": request_body.get("model"),
+        "messages": request_body.get("messages", []),
+        "stream": False,
+    }
+    if request_body.get("temperature") is not None:
+        kw["temperature"] = request_body["temperature"]
+    if request_body.get("top_p") is not None:
+        kw["top_p"] = request_body["top_p"]
+    if request_body.get("max_tokens") is not None:
+        kw["max_tokens"] = request_body["max_tokens"]
+    if request_body.get("stop") is not None:
+        kw["stop"] = request_body["stop"]
+    tools = request_body.get("tools")
+    if tools:
+        kw["tools"] = tools
+        kw["tool_choice"] = request_body.get("tool_choice", "auto")
+    return kw
+
+
+def _normalize_tool_calls(tcs):
+    if not tcs:
+        return None
+    out = []
+    for tc in tcs:
+        out.append(
+            {
+                "id": getattr(tc, "id", None),
+                "type": "function",
+                "function": {
+                    "name": getattr(tc.function, "name", None),
+                    "arguments": getattr(tc.function, "arguments", "") or "",
+                },
+            }
+        )
+    return out
+
+
+async def achat_complete(api_url, api_key, request_body, *, proxy=None, timeout=120, stream=False):
+    """统一的 OpenAI 兼容聊天补全。详见模块 docstring。"""
+    if stream:
+
+        async def _gen():
+            client = build_openai_client(api_url, api_key, proxy=proxy, timeout=timeout)
+            try:
+                gen = await client.chat.completions.create(
+                    **_create_kwargs(request_body), stream=True
+                )
+                async for chunk in gen:
+                    if not chunk.choices:
+                        continue
+                    delta = chunk.choices[0].delta
+                    rc = getattr(delta, "reasoning_content", None)
+                    if rc:
+                        yield ("reasoning", rc)
+                    ct = getattr(delta, "content", None)
+                    if ct:
+                        yield ("content", ct)
+            finally:
+                await client.close()
+
+        return _gen()
+
+    client = build_openai_client(api_url, api_key, proxy=proxy, timeout=timeout)
+    async with client:
+        resp = await client.chat.completions.create(**_create_kwargs(request_body))
+        msg = resp.choices[0].message
+        return {
+            "choices": [
+                {
+                    "message": {
+                        "content": msg.content or "",
+                        "tool_calls": _normalize_tool_calls(getattr(msg, "tool_calls", None)),
+                        "reasoning_content": getattr(msg, "reasoning_content", None),
+                    }
+                }
+            ],
+            "usage": (resp.usage.model_dump() if resp.usage else {}),
+        }
+
+
+def build_openai_client_sync(api_url, api_key, *, proxy=None, timeout=120) -> OpenAI:
+    """同步客户端工厂（供 IM bot 等同步上下文复用同一标准模型层）。"""
+    base = _base_url(api_url)
+    http_client = httpx.Client(proxy=proxy, trust_env=False)
+    return OpenAI(base_url=base, api_key=api_key or "EMPTY", timeout=timeout, http_client=http_client)
+
+
+def chat_complete(api_url, api_key, request_body, *, proxy=None, timeout=120):
+    """同步版（与 achat_complete 行为/返回形状完全一致），供非 async 上下文使用。"""
+    client = build_openai_client_sync(api_url, api_key, proxy=proxy, timeout=timeout)
+    with client:
+        resp = client.chat.completions.create(**_create_kwargs(request_body))
+        msg = resp.choices[0].message
+        return {
+            "choices": [
+                {
+                    "message": {
+                        "content": msg.content or "",
+                        "tool_calls": _normalize_tool_calls(getattr(msg, "tool_calls", None)),
+                        "reasoning_content": getattr(msg, "reasoning_content", None),
+                    }
+                }
+            ],
+            "usage": (resp.usage.model_dump() if resp.usage else {}),
+        }
+
+
+# ---------------------------------------------------------------------------
+# 嵌入（Embeddings）：OpenAI 兼容 /v1/embeddings，收敛散落的裸 httpx 嵌入调用
+# ---------------------------------------------------------------------------
+def _embed_kwargs(input, model):
+    return {"model": model, "input": input if isinstance(input, list) else [input]}
+
+
+async def aembed(api_url, api_key, input, model, *, proxy=None, timeout=60) -> list:
+    """统一的 OpenAI 兼容文本嵌入（异步）。返回 list[list[float]]。
+
+    供 pgvector_search.get_embedding 等 embeddings 路径复用，把散落的裸 httpx
+    嵌入调用收敛到同一标准模型层（与聊天共用代理/超时/failover 语义）。
+    """
+    client = build_openai_client(api_url, api_key, proxy=proxy, timeout=timeout)
+    async with client:
+        resp = await client.embeddings.create(**_embed_kwargs(input, model))
+        return [d.embedding for d in resp.data]
+
+
+def embed(api_url, api_key, input, model, *, proxy=None, timeout=60) -> list:
+    """同步版嵌入（与非 async 上下文复用同一标准模型层）。"""
+    client = build_openai_client_sync(api_url, api_key, proxy=proxy, timeout=timeout)
+    with client:
+        resp = client.embeddings.create(**_embed_kwargs(input, model))
+        return [d.embedding for d in resp.data]

backend/python/app/core/observability.py

@@ -1,6 +1,6 @@
 """
-Qihu-AIFlow 统一可观测基座 (Phase 2.1)
-=====================================
+Qihu-AIFlow 统一可观测基座 (Phase 2.1 + Phase 1.I OTel 增强)
+=============================================================
 提供贯穿一次请求的全局链路 ID（trace_id）与结构化日志能力，
 为后续重构（上下文引擎、媒体/LLM 成本归因）建立回归基线。
 
@@ -10,13 +10,17 @@ Qihu-AIFlow 统一可观测基座 (Phase 2.1)
   TraceIdFilter 自动注入，调用方无需改动。
 - 不引入第三方依赖（structlog 未安装），自带可选的 JSON formatter。
 
+Phase 1.I 增强：span() 在保持原有日志计时的同时，额外嵌套一个标准 OpenTelemetry
+span（经 app.core.otel 桥接，OTLP 导出到 Grafana）。OTel 失败时自动降级为 no-op，
+绝不改变既有日志行为与请求链路。
+
 使用方式：
     from app.core.observability import set_trace_id, get_trace_id, span
 
     # 在中间件里为每个请求注入：
     set_trace_id(request_id)
 
-    # 在关键调用处打计时 span（自动带 trace_id）：
+    # 在关键调用处打计时 span（自动带 trace_id + OTel span）：
     async with span("chat.send_message", conv_id=conv_id):
         return await _send_message_core(...)
 """
@@ -26,6 +30,8 @@ import uuid
 import json
 from contextvars import ContextVar
 
+from app.core.otel import current_span_cm, record_span_metrics
+
 # ---------------------------------------------------------------------------
 # trace_id：贯穿一次请求（含其内所有子调用/协程）的全局链路 ID
 # ---------------------------------------------------------------------------
@@ -103,10 +109,10 @@ def get_logger(name: str | None = None) -> logging.Logger:
 
 
 # ---------------------------------------------------------------------------
-# span：结构化计时上下文管理器（耗时 + trace_id + 自定义字段）
+# span：结构化计时上下文管理器（耗时 + trace_id + 自定义字段 + OTel 嵌套）
 # ---------------------------------------------------------------------------
 class _Span:
-    """同步/异步通用的计时 span。退出时记录 elapsed_ms 与状态。"""
+    """同步/异步通用的计时 span。退出时记录 elapsed_ms 与状态，并结束嵌套的 OTel span。"""
 
     def __init__(self, name: str, **fields):
         self.name = name
@@ -114,6 +120,7 @@ class _Span:
         self._logger = logging.getLogger("aiflow.span")
         self._start = None
         self.elapsed_ms = 0.0
+        self._otel_cm = None  # 嵌套的 OTel span 上下文管理器
 
     def _enter(self):
         self._start = time.perf_counter()
@@ -121,6 +128,10 @@ class _Span:
             f"SPAN_START {self.name}",
             extra={"span": self.name, "event": "start", "trace_id": get_trace_id(), **self.fields},
         )
+        # OTel 嵌套 span（失败安全：current_span_cm 返回 None 时 no-op）
+        self._otel_cm = current_span_cm(self.name, self.fields)
+        if self._otel_cm is not None:
+            self._otel_cm.__enter__()
         return self
 
     def _exit(self, exc_type, exc, tb):
@@ -137,6 +148,24 @@ class _Span:
                 **self.fields,
             },
         )
+        # 结束嵌套的 OTel span（失败安全）
+        if self._otel_cm is not None:
+            try:
+                from opentelemetry import trace as _ot
+                if exc_type is not None:
+                    self._otel_cm.set_status(
+                        _ot.trace.Status(_ot.trace.StatusCode.ERROR, str(exc_type.__name__))
+                    )
+                else:
+                    self._otel_cm.set_status(_ot.trace.Status(_ot.trace.StatusCode.OK))
+                self._otel_cm.__exit__(exc_type, exc, tb)
+            except Exception:
+                pass
+        # 记录 span 指标（耗时直方图 + 计数），供 Grafana 出图（失败安全）
+        try:
+            record_span_metrics(self.name, self.elapsed_ms, status)
+        except Exception:
+            pass
         return False  # 不吞掉异常
 
     # 同步协议

backend/python/app/core/otel.py

@@ -0,0 +1,131 @@
+"""
+OpenTelemetry 桥接 (Phase 1.I 标准化)
+=====================================
+设计基准（标准对照表见 STANDARDS_AUDIT_2026-07-24.md §I）：
+- 标准：OpenTelemetry SDK + OTLP gRPC exporter —— 云原生可观测事实标准。
+- traces：span 经 observability.span() 嵌套导出（见 observability.py）。
+- metrics：MeterProvider 经 OTLP 导出 span 耗时直方图 + 调用计数，供 Grafana 出图。
+- 失败安全（生产红线）：若 OTLP 端点不可达 / 初始化异常，自动降级为 no-op，
+  绝不阻断业务请求、不产生未捕获异常。
+- 门控：OTEL_ENABLED=1 默认开启；置 0 则零开销、不走 OTel。
+- 发射目标：OTEL_EXPORTER_OTLP_ENDPOINT，默认 http://qihu-aiflow-alloy:4317
+  （经 Grafana Alloy 桥接 OTLP→Prometheus，再由 Prometheus 喂给 Grafana 出图）。
+
+对外 API：
+    init_otel()            首次调用时构建 TracerProvider + MeterProvider（幂等）
+    get_tracer()           返回 tracer（未启用/失败则 None）
+    get_meter()            返回 meter（未启用/失败则 None）
+    current_span_cm(name)  返回一个 OTel span 的上下文管理器（供 observability 嵌套）
+    record_span_metrics(name, elapsed_ms, status)  记录 span 指标（耗时直方图 + 计数）
+"""
+
+import os
+import logging
+
+logger = logging.getLogger(__name__)
+
+_ENABLED = os.getenv("OTEL_ENABLED", "1") not in ("0", "false", "False", "")
+_ENDPOINT = os.getenv("OTEL_EXPORTER_OTLP_ENDPOINT", "http://qihu-aiflow-alloy:4317").rstrip("/")
+if not _ENDPOINT.startswith("http://") and not _ENDPOINT.startswith("https://"):
+    _ENDPOINT = "http://" + _ENDPOINT
+
+_tracer = None
+_meter = None
+_initialized = False
+_duration_hist = None
+_span_counter = None
+
+
+def init_otel() -> None:
+    """构建 TracerProvider + MeterProvider（幂等）。失败则降级为 no-op。"""
+    global _tracer, _meter, _initialized
+    if _initialized:
+        return
+    _initialized = True
+    if not _ENABLED:
+        logger.info("[otel] disabled (OTEL_ENABLED=0)")
+        return
+    try:
+        from opentelemetry import trace, metrics
+        from opentelemetry.sdk.resources import Resource
+        from opentelemetry.semconv.resource import ResourceAttributes
+        from opentelemetry.sdk.trace import TracerProvider
+        from opentelemetry.sdk.trace.export import BatchSpanProcessor
+        from opentelemetry.exporter.otlp.proto.grpc.trace_exporter import (
+            OTLPSpanExporter,
+        )
+        from opentelemetry.sdk.metrics import MeterProvider
+        from opentelemetry.sdk.metrics.export import PeriodicExportingMetricReader
+        from opentelemetry.exporter.otlp.proto.grpc.metric_exporter import (
+            OTLPMetricExporter,
+        )
+
+        resource = Resource.create({ResourceAttributes.SERVICE_NAME: "qihu-aiflow"})
+
+        # --- traces ---
+        trace_provider = TracerProvider(resource=resource)
+        trace_provider.add_span_processor(
+            BatchSpanProcessor(OTLPSpanExporter(endpoint=_ENDPOINT, insecure=True))
+        )
+        trace.set_tracer_provider(trace_provider)
+        _tracer = trace.get_tracer("qihu-aiflow")
+
+        # --- metrics ---
+        metric_reader = PeriodicExportingMetricReader(
+            OTLPMetricExporter(endpoint=_ENDPOINT, insecure=True)
+        )
+        meter_provider = MeterProvider(resource=resource, metric_readers=[metric_reader])
+        metrics.set_meter_provider(meter_provider)
+        _meter = metrics.get_meter("qihu-aiflow")
+
+        logger.info(f"[otel] tracer+meter ready -> {_ENDPOINT}")
+    except Exception as e:  # pragma: no cover - 降级保护
+        logger.warning(f"[otel] init failed, degraded to no-op: {e}")
+        _tracer = None
+        _meter = None
+
+
+def get_tracer():
+    """返回 tracer；首次调用触发 init。未启用/失败返回 None。"""
+    if _tracer is None and _ENABLED:
+        init_otel()
+    return _tracer
+
+
+def get_meter():
+    """返回 meter；首次调用触发 init。未启用/失败返回 None。"""
+    if _meter is None and _ENABLED:
+        init_otel()
+    return _meter
+
+
+def current_span_cm(name: str, attributes: dict | None = None):
+    """返回一个 OTel span 的上下文管理器（同步/异步通用）。失败时返回 None。"""
+    tr = get_tracer()
+    if tr is None:
+        return None
+    try:
+        return tr.start_as_current_span(name, attributes=attributes or {})
+    except Exception as e:  # pragma: no cover
+        logger.warning(f"[otel] start_span failed: {e}")
+        return None
+
+
+def record_span_metrics(name: str, elapsed_ms: float, status: str) -> None:
+    """记录 span 指标：耗时直方图 + 调用计数（供 Grafana 出图）。失败安全。"""
+    m = get_meter()
+    if m is None:
+        return
+    global _duration_hist, _span_counter
+    try:
+        if _duration_hist is None:
+            _duration_hist = m.create_histogram(
+                "aiflow_span_duration_ms", "Span duration in milliseconds"
+            )
+            _span_counter = m.create_counter("aiflow_span_count", "Span invocation count")
+        attrs = {"span": name, "status": status}
+        _duration_hist.record(elapsed_ms, attrs)
+        _span_counter.add(1, attrs)
+    except Exception:  # pragma: no cover
+        pass
+

backend/python/app/core/rag_pipeline.py

@@ -0,0 +1,97 @@
+"""标准 RAG 管线 (Phase 2.C 标准化)
+====================================
+设计基准（标准对照表见 STANDARDS_AUDIT_2026-07-24.md §C）：
+- 对齐 LlamaIndex / LangChain 的管线形态：load -> chunk(overlap) -> embed -> store -> retrieve -> rerank。
+- 复用现有 pgvector 后端（document_embeddings 表 + ivfflat）与共享 embeddings 客户端
+  （app.core.llm_client.aembed）；嵌入维度(1024)、表结构、混合检索(RRF)全部沿用，不重复造轮子。
+- 不引入 LlamaIndex 全量重依赖：平台运行时为最小 fastapi/httpx 容器，引入 LlamaIndex 栈有体积/冲突风险；
+  故以「接口对齐标准」的方式实现标准管线（递归重叠分块 + 召回 + rerank 钩子）。后续若评估可接受，
+  可平滑替换为 llama_index.core 的 VectorStoreIndex + PGVectorStore（对外接口一致）。
+
+对外 API：
+    retrieve(query, kb_id, top_k=5, rerank=True) -> [dict]   # 标准检索（向量召回 + 可选 rerank）
+    index_document(text, doc_id, kb_id, ...) -> int          # 标准索引（递归重叠分块后逐块入库）
+"""
+import logging
+
+from app.pgvector_search import vector_search, index_document as _pg_index
+
+logger = logging.getLogger("aiflow.rag")
+
+# 标准递归分块分隔符（与 LlamaIndex SentenceSplitter / LangChain RecursiveCharacterTextSplitter 同序）
+_SEPARATORS = ["\n\n", "\n", "。", "！", "？", "；", ".", "!", "?", ";", " ", ""]
+
+
+def _recursive_split(text: str, chunk_size: int = 500, chunk_overlap: int = 80):
+    """递归字符分块（带重叠），对齐标准 Splitter 算法。
+
+    逐层用分隔符切分；某段仍超长时递归细切；最终兜底硬切（带重叠步长）。
+    """
+    text = text or ""
+    if len(text) <= chunk_size:
+        return [text] if text.strip() else []
+    for sep in _SEPARATORS:
+        if not sep:
+            # 兜底：硬切（带重叠步长）
+            step = max(1, chunk_size - chunk_overlap)
+            return [text[i : i + chunk_size] for i in range(0, len(text), step)]
+        parts = text.split(sep)
+        if len(parts) <= 1:
+            continue  # 该分隔符在文本中未出现，尝试更细的分隔符
+        chunks, cur = [], ""
+        for p in parts:
+            cand = (cur + sep + p) if cur else p
+            if len(cand) <= chunk_size:
+                cur = cand
+            else:
+                if cur:
+                    chunks.append(cur)
+                    cur = ""
+                if len(p) > chunk_size:
+                    chunks.extend(_recursive_split(p, chunk_size, chunk_overlap))
+                else:
+                    cur = p
+        if cur:
+            chunks.append(cur)
+        if chunks:
+            return [c for c in chunks if c.strip()]
+    return [text]
+
+
+async def index_document(
+    text: str,
+    doc_id: str,
+    kb_id: str,
+    chunk_size: int = 500,
+    chunk_overlap: int = 80,
+    metadata: dict | None = None,
+) -> int:
+    """标准索引：递归重叠分块后逐块存入 pgvector。返回成功分块数。"""
+    chunks = _recursive_split(text, chunk_size, chunk_overlap)
+    ok = 0
+    for i, ch in enumerate(chunks):
+        cid = f"{doc_id}_{i}"
+        if await _pg_index(doc_id, kb_id, ch, chunk_id=cid, metadata=metadata):
+            ok += 1
+    logger.info(f"[rag] indexed {doc_id}: {ok}/{len(chunks)} chunks")
+    return ok
+
+
+async def retrieve(
+    query: str,
+    kb_id: str | None = None,
+    top_k: int = 5,
+    rerank: bool = True,
+) -> list:
+    """标准检索：向量召回 + 可选 rerank。
+
+    返回 [{doc_id, chunk_id, content, score, metadata}, ...]。
+    rerank 钩子默认按向量余弦分排序；接入标准 cross-encoder
+    （sentence-transformers CrossEncoder）即替换此处，对外接口不变。
+    """
+    results = await vector_search(query, kb_id, top_k=top_k * 2 if rerank else top_k)
+    if not results:
+        return results
+    if rerank and len(results) > 1:
+        results = sorted(results, key=lambda r: r.get("score", 0), reverse=True)
+    return results[:top_k]