feat: 完成 issue #133 对话助手增强版（引用溯源/路由分级/DLP 拦截，直连推理服务）

2026-08-05 09:57:53 +08:00
parent 19c832b244
commit faddf24d11
6 changed files with 602 additions and 0 deletions
@@ -0,0 +1,80 @@
+# -*- coding: utf-8 -*-
+"""对话助手 RAG 引用语料编译脚本（issue #133 / PRD 5.4 引用溯源）。
+
+扫描行业模板知识库文档（``templates/*/rag-kb/documents/*.md``），抽取
+「文档标题 + 章节 + 章节摘要」编译为 ``web/chat/citations.json``，
+供增强版对话页（assistant.html）做前端检索与**引用溯源展示**。
+
+与 rag-kb 内核的关系：``core/rag-kb`` 是完整检索管线（暂无 HTTP 服务），
+本脚本只把文档资产转成前端可 fetch 的静态语料（Demo 级），
+引用条目带「文档名 + 章节」，对齐 PRD 5.4「强制引用溯源」的展示语义。
+
+用法（仓库根目录下）：
+    python web/chat/scripts/build_citations.py
+"""
+from __future__ import annotations
+
+import glob
+import json
+import os
+import re
+import sys
+
+_REPO_ROOT = os.path.dirname(
+    os.path.dirname(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
+)
+_DOC_GLOB = os.path.join(_REPO_ROOT, "templates", "*", "rag-kb", "documents", "*.md")
+_OUT_PATH = os.path.join(_REPO_ROOT, "web", "chat", "citations.json")
+
+#: 每个章节摘要的最大字符数（控制语料体积，前端检索只看要点）
+SNIPPET_MAX_CHARS = 160
+
+
+def extract_sections(path: str) -> dict:
+    """从一份 Markdown 文档抽取标题与章节摘要。"""
+    with open(path, encoding="utf-8") as f:
+        text = f.read()
+    rel = os.path.relpath(path, _REPO_ROOT).replace(os.sep, "/")
+    title_match = re.search(r"^#\s+(.+)$", text, flags=re.M)
+    title = title_match.group(1).strip() if title_match else os.path.basename(path)
+
+    sections = []
+    # 按 ## 章节切分；无章节时整篇作为一节
+    parts = re.split(r"^##\s+(.+)$", text, flags=re.M)
+    if len(parts) >= 3:
+        for i in range(1, len(parts), 2):
+            heading = parts[i].strip()
+            body = parts[i + 1] if i + 1 < len(parts) else ""
+            snippet = re.sub(r"\s+", " ", body).strip()[:SNIPPET_MAX_CHARS]
+            if snippet:
+                sections.append({"heading": heading, "snippet": snippet})
+    else:
+        snippet = re.sub(r"\s+", " ", text).strip()[:SNIPPET_MAX_CHARS]
+        if snippet:
+            sections.append({"heading": "全文", "snippet": snippet})
+
+    return {"title": title, "source": rel, "sections": sections}
+
+
+def main() -> int:
+    docs = []
+    for path in sorted(glob.glob(_DOC_GLOB)):
+        try:
+            docs.append(extract_sections(path))
+            print(f"[OK] {os.path.basename(path)} "
+                  f"({len(docs[-1]['sections'])} 章节)")
+        except Exception as exc:  # noqa: BLE001 - 聚合全部失败
+            print(f"[FAIL] {path}: {exc}")
+            return 1
+    if not docs:
+        print("未找到任何知识库文档")
+        return 1
+    with open(_OUT_PATH, "w", encoding="utf-8") as f:
+        json.dump({"schema": "iAOP-chat-citations-v1", "documents": docs},
+                  f, ensure_ascii=False, indent=2)
+    print(f"共 {len(docs)} 份文档 -> {os.path.relpath(_OUT_PATH, _REPO_ROOT)}")
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())