账号支持只看本账号看板；模板重新解析改为保留原映射并多轮补缺。

Co-authored-by: Cursor <cursoragent@cursor.com>

inquiry-agent/agent/channel/quote_template_parse.py

@@ -37,6 +37,9 @@ class QuoteTemplateParseBody(BaseModel):
     yellow_count: int = Field(default=0, alias="yellowCount")
     truncated: bool = False
     request_id: str = Field(default="", alias="requestId")
+    # 未换文件重新解析：true + existingMapping，回写时旧字段优先补缺
+    merge_existing: bool = Field(default=False, alias="mergeExisting")
+    existing_mapping: Optional[dict[str, Any]] = Field(default=None, alias="existingMapping")
 
     model_config = {"populate_by_name": True}
 
@@ -103,7 +106,10 @@ def enqueue_parse(
         "cellCount": body.cell_count or len(inventory),
         "yellowCount": body.yellow_count,
         "truncated": bool(body.truncated),
+        "mergeExisting": bool(body.merge_existing),
     }
+    if body.merge_existing and isinstance(body.existing_mapping, dict):
+        payload["existingMapping"] = body.existing_mapping
     result = enqueue_recognition(
         kind="quote_template_parse",
         payload=payload,

inquiry-agent/agent/jobs/recognition.py

@@ -111,28 +111,50 @@ def process_recognition(data: dict[str, Any], *, allow_network: bool = True) ->
 
 
 def _process_quote_template_parse(payload: dict[str, Any], *, allow_network: bool) -> dict[str, Any]:
-    """跑 sheet_map 并回写主账；失败也必须回写 failed+原因。"""
+    """
+    跑 sheet_map 并回写主账；失败也必须回写 failed+原因。
+
+    mergeExisting=true（未换文件重新解析）时：与 existingMapping 补缺合并后再回写，
+    旧字段优先保留；换文件/首传不传该标记，整份覆盖。
+    """
     from agent.config import get_settings
-    from agent.llm.mode_sheet_map import invoke_sheet_map
+    from agent.llm.mode_sheet_map import (
+        assess_mapping_gaps,
+        invoke_sheet_map,
+        merge_mapping_prefer_existing,
+    )
 
     version_id = str(payload.get("versionId") or "").strip()
     settings = get_settings()
     net = bool(allow_network and settings.llm_allow_network)
-    result = invoke_sheet_map(payload, allow_network=net)
+    result = invoke_sheet_map(payload, allow_network=net, timeout_seconds=240.0, max_rounds=3)
+    merge_existing = bool(payload.get("mergeExisting") or payload.get("merge_existing"))
+    existing = payload.get("existingMapping") or payload.get("existing_mapping")
     body: dict[str, Any]
     if result.get("ok") and isinstance(result.get("mapping"), dict):
         mapping = dict(result["mapping"])
-        gaps = result.get("gaps_after")
-        if not isinstance(gaps, list):
-            gaps = mapping.get("mappingGaps") if isinstance(mapping.get("mappingGaps"), list) else []
-        mapping["mappingGaps"] = [str(g) for g in gaps if g]
+        if merge_existing and isinstance(existing, dict):
+            mapping = merge_mapping_prefer_existing(existing, mapping)
+            # 合并后再对照清单复评缺口，避免角标沿用合并前数字
+            remaining = assess_mapping_gaps(payload, mapping)
+            mapping["mappingGaps"] = [str(g) for g in remaining if g]
+            if not remaining:
+                mapping["unmappedRequired"] = []
+                mapping["unmappedLabels"] = []
+        else:
+            gaps = result.get("gaps_after")
+            if not isinstance(gaps, list):
+                gaps = mapping.get("mappingGaps") if isinstance(mapping.get("mappingGaps"), list) else []
+            mapping["mappingGaps"] = [str(g) for g in gaps if g]
         # 运营可读摘要：点数之外固定带上未解析项
         n_fields = len(mapping.get("fields") or {})
         n_fee = len(mapping.get("feeRegions") or []) if isinstance(mapping.get("feeRegions"), list) else 0
         n_route = len(mapping.get("routeRegions") or []) if isinstance(mapping.get("routeRegions"), list) else 0
         points = n_fields + n_fee + n_route
         parse_message = f"已解析 {points} 个映射点"
-        if mapping["mappingGaps"]:
+        if merge_existing:
+            parse_message += "（保留原映射并补缺）"
+        if mapping.get("mappingGaps"):
             preview = "；".join(mapping["mappingGaps"][:8])
             more = len(mapping["mappingGaps"]) - 8
             parse_message += f"；未解析到 {len(mapping['mappingGaps'])} 项：{preview}"
@@ -144,6 +166,7 @@ def _process_quote_template_parse(payload: dict[str, Any], *, allow_network: boo
             "mappingJson": mapping,
             "parseStatus": "parsed",
             "parseMessage": parse_message,
+            "mergeExisting": merge_existing,
         }
     else:
         err = str(result.get("error") or "AI 解析失败")
@@ -152,12 +175,14 @@ def _process_quote_template_parse(payload: dict[str, Any], *, allow_network: boo
             "ok": False,
             "parseStatus": "failed",
             "parseMessage": err,
+            "mergeExisting": merge_existing,
         }
     posted = _post_parse_result(body)
     logger.info(
-        "quote_template_parse done versionId=%s ok=%s posted=%s",
+        "quote_template_parse done versionId=%s ok=%s merge=%s posted=%s",
         version_id,
         body.get("ok"),
+        merge_existing,
         posted.get("ok"),
     )
     return {"ok": bool(body.get("ok")), "posted": posted, "kind": "quote_template_parse"}

inquiry-agent/agent/llm/mode_sheet_map.py

@@ -2,7 +2,7 @@
 报价工作表映射 mode：上传时用千问产出 mappingJson。
 
 本文件职责：加载映射分析提示词、拼用户消息、调千问、剥 Markdown、校验 JSON、
-对照单元格清单做缺项检测，必要时再补跑一轮。
+对照单元格清单做缺项检测；有缺口则多轮补跑（默认最多 3 轮），目标一次流程内齐全。
 禁止：发明运价；在企微回调线程同步调用；出单阶段再调本 mode。
 """
 
@@ -22,30 +22,58 @@ MODE = LlmMode.SHEET_MAP
 
 _PROMPT_CACHE: str | None = None
 
-# 清单原文片段 → 白名单 code（用于缺项检测；不要求全覆盖词典）
-_LABEL_TO_CODE: list[tuple[tuple[str, ...], str]] = [
-    (("提货地址", "Place of Receipt", "place of receipt"), "pickup_address"),
-    (("送货地址", "派送地址", "Place of Delivery", "place of delivery"), "delivery_address"),
-    (("货值", "Cargo Value", "cargo value"), "cargo_value"),
-    (("海关编码", "HS Code", "HS编码", "hs code"), "hs_code"),
-    (("油、电、磁", "油,电,磁", "油电磁", "IL, ELEC", "L,ELEC", "oil"), "oil_elec_mag"),
+# 清单原文片段 → 白名单 code（按运输方式过滤；关键词避免过短英文误报）
+_LABEL_TO_CODE_COMMON: list[tuple[tuple[str, ...], str]] = [
+    (("报价单编号", "询价单号", "RFQ No", "Quote No"), "quote_no"),
+    (("报价日期", "询价日期", "Inquiry Date", "Quote Date"), "quote_date"),
+    (("有效期至", "有效期", "Validity"), "valid_until"),
+    (("客户名称", "Customer Name", "Customer"), "customer_name"),
+    (("联系人", "Contact Person", "Contact Name"), "contact_name"),
+    (("电话", "Tel", "Phone", "Mobile"), "contact_phone"),
+    (("币别", "Currency"), "currency"),
+    (("业务员", "Sales"), "sales_name"),
+    (("品名", "Commodity", "Goods Description"), "commodity"),
+]
+
+_LABEL_TO_CODE_SEA: list[tuple[tuple[str, ...], str]] = [
+    (("提货地址", "Place of Receipt"), "pickup_address"),
+    (("送货地址", "派送地址", "Place of Delivery"), "delivery_address"),
+    (("货值", "Cargo Value"), "cargo_value"),
+    (("海关编码", "HS Code", "HS编码"), "hs_code"),
+    (("油、电、磁", "油,电,磁", "油电磁"), "oil_elec_mag"),
     (("特殊说明", "特别说明", "Special Remarks"), "special_remarks"),
-    (("要求货物到达", "Requested Arriv", "抵达时间"), "eta_request"),
-    (("运输时效", "Transit Time", "航程时间", "Navigation time"), "transit_time"),
-    (("船期", "Sailing", "Requested Sh"), "sailing_schedule"),
-    (("开船日", "On Board"), "on_board_date"),
-    (("船公司", "Shipping Line", "Shipping Co"), "shipping_line"),
+    (("要求货物到达", "Requested Arrival"), "eta_request"),
+    (("运输时效", "Transit Time", "航程时间"), "transit_time"),
+    (("船期", "Sailing Schedule"), "sailing_schedule"),
+    (("开船日", "On Board Date"), "on_board_date"),
+    (("船公司", "Shipping Line", "Shipping Company"), "shipping_line"),
     (("货好时间", "Cargo Ready"), "cargo_ready_date"),
-    (("贸易条款", "Terms"), "trade_terms"),
+    (("贸易条款", "Trade Terms", "Incoterms"), "trade_terms"),
     (("整柜或散货", "整柜或拼柜", "FCL or LCL"), "fcl_or_lcl"),
-    (("箱型", "柜量", "Type & Qty", "Cont."), "container_type_qty"),
-    (("起运港", "POL"), "pol"),
-    (("目的港", "POD"), "pod"),
-    (("品名", "Commodity"), "commodity"),
+    (("箱型箱量", "箱型", "柜量", "Type & Qty"), "container_type_qty"),
+    (("起运港", "Port of Loading", "POL"), "pol"),
+    (("目的港", "Port of Discharge", "POD"), "pod"),
     (("货物数量", "Quantity"), "quantity"),
-    (("询价日期", "Inquiry Date", "报价日期"), "quote_date"),
-    (("RFQ", "询价单号", "报价单编号"), "quote_no"),
-    (("有效期", "period"), "valid_until"),
+]
+
+_LABEL_TO_CODE_AIR: list[tuple[tuple[str, ...], str]] = [
+    (("起运地", "起运港", "Departure", "Origin Airport"), "origin"),
+    (("目的地", "目的港", "Destination Airport"), "destination"),
+    (("件数", "Pieces", "Pkgs"), "pieces"),
+    (("毛重", "重量", "Gross Weight", "Weight"), "weight_kg"),
+    (("体积", "Volume", "CBM"), "volume_cbm"),
+    (("计费重", "Chargeable Weight"), "chargeable_weight_kg"),
+    (("航班", "Flight"), "flight_no"),
+    (("中转机场", "Transfer Airport"), "transfer_airport"),
+    (("贸易条款", "Trade Terms", "Incoterms"), "trade_terms"),
+    (("特殊说明", "特别说明", "Special Remarks"), "special_remarks"),
+]
+
+_LABEL_TO_CODE_LAND: list[tuple[tuple[str, ...], str]] = [
+    (("始发地", "起运地", "发站", "Origin"), "origin"),
+    (("目的地", "到站", "Destination"), "destination"),
+    (("运输时效", "时效", "Transit Time"), "transit_time"),
+    (("特殊说明", "特别说明"), "special_remarks"),
 ]
 
 _SEA_FEE_MARKERS: list[tuple[tuple[str, ...], tuple[str, ...]]] = [
@@ -65,6 +93,33 @@ _SEA_FEE_MARKERS: list[tuple[tuple[str, ...], tuple[str, ...]]] = [
 ]
 
 
+def _labels_for_mode(mode: str) -> list[tuple[tuple[str, ...], str]]:
+    m = (mode or "sea").strip().lower()
+    out = list(_LABEL_TO_CODE_COMMON)
+    if m in ("sea", "海运"):
+        out.extend(_LABEL_TO_CODE_SEA)
+    elif m in ("air", "空运"):
+        out.extend(_LABEL_TO_CODE_AIR)
+    elif m in ("land", "陆运"):
+        out.extend(_LABEL_TO_CODE_LAND)
+    else:
+        out.extend(_LABEL_TO_CODE_SEA)
+    return out
+
+
+def _label_hit(blob: str, labels: tuple[str, ...]) -> bool:
+    """清单命中：中文标签直接包含；英文至少 4 字符，降低误报。"""
+    for lab in labels:
+        if not lab:
+            continue
+        if any("\u4e00" <= ch <= "\u9fff" for ch in lab):
+            if lab in blob:
+                return True
+        elif len(lab) >= 4 and lab.lower() in blob.lower():
+            return True
+    return False
+
+
 def mode_meta() -> dict[str, Any]:
     return {
         "mode": MODE.value,
@@ -178,7 +233,7 @@ def assess_mapping_gaps(payload: dict[str, Any], mapping: dict[str, Any] | None)
     """
     对照单元格清单，列出 mapping 明显缺口（中文说明，供补跑用户消息）。
 
-    不打外网。海运：费用段标题出现则要求对应 feeRegion；常见业务标签出现则要求 fields code。
+    不打外网。按运输方式过滤字段词典；海运另检查三段费用区。
     """
     blob = _inventory_blob(payload)
     if not blob.strip():
@@ -196,24 +251,89 @@ def assess_mapping_gaps(payload: dict[str, Any], mapping: dict[str, Any] | None)
         for r in fee_regions
         if isinstance(r, dict)
     ).lower()
+    route_regions = mapping.get("routeRegions") if isinstance(mapping.get("routeRegions"), list) else []
 
     gaps: list[str] = []
     mode = str(payload.get("transportMode") or payload.get("transport_mode") or "sea").strip().lower()
+    # 字段匹配用：去掉费用段标题，避免「目的港费用」误报成 pod 等字段缺口
+    field_blob = blob
     if mode in ("sea", "海运"):
         for markers, matchers in _SEA_FEE_MARKERS:
             if not any(m in blob for m in markers):
                 continue
             if not any(x in fee_blob for x in matchers):
                 gaps.append(f"费用区缺失：清单有「{markers[0]}」，须建对应 feeRegion")
+            for m in markers:
+                field_blob = field_blob.replace(m, " ")
+    elif mode in ("land", "陆运"):
+        # 陆运主表常见项目/线路矩阵：清单有「项目」或车型价列但无 routeRegions
+        if (
+            ("项目" in blob or "车型" in blob or "3T" in blob or "拼车" in blob)
+            and not route_regions
+        ):
+            gaps.append("线路区缺失：陆运表须建 routeRegions 承接线路/车型报价")
 
-    for labels, code in _LABEL_TO_CODE:
-        if code in field_codes:
+    # 陆运已建 routeRegions 时，始发地/目的地/时效常在线路矩阵列头，不再强制 fields
+    land_route_covered = mode in ("land", "陆运") and bool(route_regions)
+    land_skip_codes = {"origin", "destination", "transit_time"} if land_route_covered else set()
+
+    for labels, code in _labels_for_mode(mode):
+        if code in field_codes or code in land_skip_codes:
             continue
-        if any(lab in blob for lab in labels):
+        if _label_hit(field_blob, labels):
             gaps.append(f"字段缺失：清单有「{labels[0]}」→ 应映射 {code}")
     return gaps
 
 
+def merge_mapping_prefer_existing(
+    old_mapping: dict[str, Any] | None,
+    new_mapping: dict[str, Any] | None,
+) -> dict[str, Any]:
+    """
+    补缺合并：旧 fields / 区域优先，新结果只补充旧里没有的 code/id。
+
+    用于「未换文件重新解析」：避免已对上的字段被新一轮 AI 冲掉。
+    """
+    new_mapping = dict(new_mapping or {})
+    old_mapping = old_mapping if isinstance(old_mapping, dict) else {}
+    out = dict(new_mapping)
+
+    new_fields = new_mapping.get("fields") if isinstance(new_mapping.get("fields"), dict) else {}
+    old_fields = old_mapping.get("fields") if isinstance(old_mapping.get("fields"), dict) else {}
+    fields: dict[str, Any] = dict(new_fields)
+    for code, cell in old_fields.items():
+        if cell is not None and str(cell).strip():
+            fields[str(code)] = cell
+    out["fields"] = fields
+
+    out["feeRegions"] = _merge_regions_by_id(
+        old_mapping.get("feeRegions"), new_mapping.get("feeRegions")
+    )
+    out["routeRegions"] = _merge_regions_by_id(
+        old_mapping.get("routeRegions"), new_mapping.get("routeRegions")
+    )
+    return out
+
+
+def _merge_regions_by_id(old_arr: Any, new_arr: Any) -> list[dict[str, Any]]:
+    """区域按 id 合并：旧整段保留，新 id 追加。"""
+    out: list[dict[str, Any]] = []
+    seen: set[str] = set()
+    for src in (old_arr, new_arr):
+        if not isinstance(src, list):
+            continue
+        for item in src:
+            if not isinstance(item, dict):
+                continue
+            rid = str(item.get("id") or "").strip()
+            if rid and rid in seen:
+                continue
+            if rid:
+                seen.add(rid)
+            out.append(dict(item))
+    return out
+
+
 def invoke_shell(*, payload: dict[str, Any]) -> dict[str, Any]:
     """
     无网/单测入口：若带 assistant_json 则直接解析；否则返回未实现。
@@ -273,10 +393,12 @@ def invoke_sheet_map(
     *,
     allow_network: bool = False,
     chat_fn: Optional[Any] = None,
-    timeout_seconds: float = 180.0,
+    timeout_seconds: float = 240.0,
+    max_rounds: int = 3,
 ) -> dict[str, Any]:
     """
-    调用千问产出 mappingJson；若对照清单有明显缺口，再补跑一轮（最多 2 次 chat）。
+    调用千问产出 mappingJson；对照清单有缺口则补跑，最多 max_rounds 轮（默认 3），
+    目标一次流程内尽量齐全，缺口写进 mappingGaps。
 
     allow_network=False 且无 chat_fn：不打外网。
     副作用：可能出站 HTTPS；应在 Worker HTTP 槽内调用。
@@ -293,36 +415,54 @@ def invoke_sheet_map(
     else:
         chat = chat_fn
 
+    rounds_cap = max(1, min(int(max_rounds or 3), 4))
     first = _chat_once(chat=chat, payload=payload, gaps=None)
     if not first.get("ok"):
         return first
 
     mapping = first["mapping"]
-    gaps = assess_mapping_gaps(payload, mapping)
     rounds = 1
-    if gaps:
-        logger.info(
-            "sheet_map gaps versionId=%s n=%s sample=%s",
-            payload.get("versionId"),
-            len(gaps),
-            gaps[:5],
-        )
-        second = _chat_once(chat=chat, payload=payload, gaps=gaps)
-        rounds = 2
-        if second.get("ok") and second.get("mapping"):
-            mapping = second["mapping"]
-            # 第二轮仍失败结构时保留第一轮
-        elif not second.get("ok"):
-            logger.warning(
-                "sheet_map 补跑失败，保留第一轮 versionId=%s err=%s",
-                payload.get("versionId"),
-                second.get("error"),
-            )
-
     remaining = assess_mapping_gaps(payload, mapping)
-    # 缺口落进 mappingJson，主账/后台可直接展示「表上有、未映射上」
+    # 有缺口就补跑，直到齐、轮次用尽、或本轮未减少缺口
+    while remaining and rounds < rounds_cap:
+        logger.info(
+            "sheet_map gaps versionId=%s round=%s n=%s sample=%s",
+            payload.get("versionId"),
+            rounds,
+            len(remaining),
+            remaining[:5],
+        )
+        nxt = _chat_once(chat=chat, payload=payload, gaps=remaining)
+        rounds += 1
+        if not (nxt.get("ok") and nxt.get("mapping")):
+            logger.warning(
+                "sheet_map 补跑失败，保留上一轮 versionId=%s err=%s",
+                payload.get("versionId"),
+                nxt.get("error"),
+            )
+            break
+        mapping = nxt["mapping"]
+        after = assess_mapping_gaps(payload, mapping)
+        if len(after) >= len(remaining):
+            # 未继续收敛则停止，避免空转烧额度
+            logger.info(
+                "sheet_map gaps 未收敛 versionId=%s before=%s after=%s stop",
+                payload.get("versionId"),
+                len(remaining),
+                len(after),
+            )
+            remaining = after
+            break
+        remaining = after
+
+    # 缺口落进 mappingJson；齐全时清空 AI 自报未映射，避免列表角标被说明性标签抬高
     if isinstance(mapping, dict):
         mapping["mappingGaps"] = list(remaining)
+        if not remaining:
+            mapping["unmappedRequired"] = []
+            mapping["unmappedLabels"] = []
+        elif not isinstance(mapping.get("unmappedLabels"), list):
+            mapping["unmappedLabels"] = []
     logger.info(
         "sheet_map ok versionId=%s rounds=%s fields=%s feeRegions=%s routeRegions=%s remain_gaps=%s",
         payload.get("versionId"),

inquiry-agent/agent/quote_templates/prompts/mapping_analyzer.txt

@@ -360,6 +360,8 @@ transportMode：sea | air | land
 5. 空运必须能承接 departureCharges + 空运费；陆运主表必须用 routeRegions 承接 landSegments。
 6. 看不清 → unmappedRequired / unmappedLabels；表上有、能对上白名单却未映射 → 视为错误，必须补进 fields。
 7. 用户未声明运输方式时推断 transportMode，并在 notes 说明。
+8. **一次齐全**：第一轮就要尽量映射清单中全部可见业务标签；宁可多写 fields/feeRegions/routeRegions，也不要省略。补跑时输出完整 mappingJson（覆盖式），不要只回增量。
+9. 说明性/公式性文字（如「1CBM=***KG」「重量体积取大」）无法落白名单字段时写入 unmappedLabels，不要假装成业务字段缺口。
 
 【用户消息将提供】
 运输方式：海运 | 空运 | 陆运
@@ -368,4 +370,4 @@ transportMode：sea | air | land
 （可选）TMS 样例或出单投影说明——仅对齐字段
 （补跑时）缺口清单：请据此补全整份 mappingJson，不要只回增量。
 
-输出前自检 fields 与 feeRegions/routeRegions 是否覆盖清单中的可见业务标签，再只输出 mappingJson。
+输出前自检：fields 与 feeRegions/routeRegions 是否覆盖清单中的可见业务标签；有缺口则先补齐再输出。只输出 mappingJson。