表助手的背景在宽表上不至于太长：这张表的目录最多列 600 项（表级的项、关联关系在前），其余写明还有几项没列出；导入时界面直接交来的文件原文（JSON 文字）也认，不是 JSON 的 422；导入冲突说明里的状态用界面的叫法

backend/app/data/catalog_assistant.py

@@ -79,9 +79,11 @@ TOOL_CHARS = 12_000
 RETRY_DELAYS_S = (3.0, 10.0)
 #: 等模型的时候每隔几秒发一行 SSE 注释
 HEARTBEAT_S = 10.0
-#: 背景里「目录里还没有中文名和含义的列」最多列几个；别的表指过来的关联关系最多列几条
+#: 背景里「目录里还没有中文名和含义的列」最多列几个；别的表指过来的关联关系最多列几条；这张表的目录最多列几项
+#: （几百列的宽表全部起草过有上千项，全列出来提示词就太长了：表级的项、关联关系在前，列的项在后）
 UNNAMED_SHOWN = 30
 INCOMING_SHOWN = 30
+CATALOG_ITEMS_SHOWN = 600
 
 QUERY_TOOL = "run_query"
 PROPOSE_TOOL = "propose_changes"
@@ -166,6 +168,7 @@ ASSISTANT_PROMPT_LABELS = {
     "catalog_empty": "这张表还没有数据目录。",
     "item": "- {path}（{label}）：{value}〔{source}，{status}〕",
     "item_note": "  备注：{text}",
+    "catalog_more": "（还有 {n} 项没有列出：这张表的目录太长。要改的项照上面的路径写法提，核对时按当前目录算）",
     "relation": "{columns} → {to_table}.{to_columns}{extra}",
     "relation_extra": "（{text}）",
     "cardinality": "基数 {value}",
@@ -357,10 +360,14 @@ def _catalog_lines(notes: Mapping[str, Any] | None, masked: set[str]) -> list[st
     （剖析的备注里有取值范围）。"""
     labels = ASSISTANT_PROMPT_LABELS
     lines: list[str] = []
-    for key, item in catalog._slots(notes).items():
+    # 表级的项、关联关系在前，列的项在后（稳定排序，各自照目录里的顺序）；遮罩列上的码值不算
+    slots = [(key, item) for key, item in sorted(catalog._slots(notes).items(), key=lambda kv: kv[0][0] == "c")
+             if not (key[0] == "c" and key[2] == "codes" and key[1].lower() in masked)]
+    for shown, (key, item) in enumerate(slots):
+        if shown >= CATALOG_ITEMS_SHOWN:
+            lines.append(labels["catalog_more"].format(n=len(slots) - shown))
+            break
         hidden = _touches_masked(key, item, masked)
-        if hidden and key[0] == "c" and key[2] == "codes":
-            continue
         if key[0] == "r":
             label, value = labels["field_relation"], _relation_line(item)
         else:

backend/app/data/catalog_transfer.py

@@ -30,6 +30,7 @@ tables: {表名: {notes, version}}}，含全部状态的项，表结构里已经
 from __future__ import annotations
 
 import copy
+import json
 import re
 from collections.abc import Mapping
 from dataclasses import dataclass, field
@@ -51,7 +52,8 @@ _UNNAMED = "数据源"
 
 #: 状态：照带的、只填空缺的
 _LOCKED = ("confirmed", "rejected")
-_STATUS_LABEL = {"proposed": "推断", "verified": "有确证", "confirmed": "已确认", "rejected": "已驳回"}
+#: 状态的界面叫法（同 frontend/src/lib/terms.ts 的 CATALOG_STATUS_LABEL）
+_STATUS_LABEL = {"proposed": "推断", "verified": "已验证", "confirmed": "已确认", "rejected": "已驳回"}
 
 #: 写入一直撞上别人刚改过（重读重算三次仍冲突）
 WRITE_CONFLICT_ERROR = "这张表的数据目录正被其他人修改，没有导入，请稍后重新导入"
@@ -106,7 +108,12 @@ class ImportFile:
 
 
 def read_file(content: Any, *, include_guidance: bool) -> ImportFile:
-    """导出的那个对象 → ImportFile。格式不对抛 TransferInvalid（说法给人看）。"""
+    """导出的那个对象 → ImportFile。界面直接交来文件的原文（一段 JSON 文字）也认。格式不对抛 TransferInvalid（说法给人看）。"""
+    if isinstance(content, str):
+        try:
+            content = json.loads(content)
+        except ValueError as e:
+            raise TransferInvalid("文件内容不是合法的 JSON，请选择从 AgentLab 导出的数据目录文件") from e
     if not isinstance(content, Mapping):
         raise TransferInvalid("文件内容不是数据目录导出的格式，请选择从 AgentLab 导出的数据目录文件")
     if content.get("format") != FORMAT:

backend/tests/test_catalog_assistant.py

@@ -351,6 +351,18 @@ async def test_background_has_structure_catalog_and_hides_masked_values(client,
     assert "没开剖析" not in system
 
 
+async def test_long_catalogs_are_cut_with_a_note(client, tmp_path, monkeypatch):
+    sid = await _source(client, tmp_path)
+    monkeypatch.setattr(catalog_assistant, "CATALOG_ITEMS_SHOWN", 2)
+    model = Scripted([Turn(text=["好。"])])
+    use(monkeypatch, model)
+    await _talk(client, sid, _ask("看看"))
+    system = model.seen[0][0].content
+    # 表级的项在前；遮罩列的码值本来就不算；剩下的说明还有几项
+    assert "- label（中文名）：订单" in system and "- grain（粒度）" in system and "- kind（表类型）" not in system
+    assert "还有 3 项没有列出" in system
+
+
 async def test_without_profiling_there_is_no_query_tool(client, tmp_path, monkeypatch):
     sid = await _source(client, tmp_path, profile=False)
     model = Scripted([Turn(text=["只能根据表结构推断。"])])

backend/tests/test_catalog_transfer.py

@@ -334,6 +334,11 @@ async def test_import_waits_for_draft_jobs_and_rejects_bad_files(client):
         r = await client.post(f"{_base(local)}/import/preview", json={"content": body})
         assert r.status_code == 422, body
         assert any("一" <= ch <= "鿿" for ch in r.json()["detail"])
+    # 界面直接交来文件原文也认；不是 JSON 的 422
+    as_text = await client.post(f"{_base(local)}/import/preview", json={"content": json.dumps(content)})
+    assert as_text.status_code == 200 and as_text.json()["totals"]["tables"] == 2
+    not_json = await client.post(f"{_base(local)}/import/preview", json={"content": "{不是 JSON"})
+    assert not_json.status_code == 422 and "JSON" in not_json.json()["detail"]
     too_long = {**content, "guidance": "长" * (catalog_jobs.GUIDANCE_MAX + 1)}
     r = await client.post(f"{_base(local)}/import/preview", json={"content": too_long, "include_guidance": True})
     assert r.status_code == 422 and "起草要求" in r.json()["detail"]