feat(vault): Phase 3 tools + batch citation/relation enrichment

tools/data/home-wiki-summaries.json

@@ -33,6 +33,10 @@
   "价值驱动": {
     "summary": "郭朝晖：以终为始在数字化场景的落地（手段≠目的/投入产出比/口号陷阱）"
   },
+  "军事系统工程": {
+    "summary": "来源：钱学森+王寿云+柴本良 1979 年 7 月 24 日解放军总部机关领导同志学习会讲演稿 钱学森主讲，王寿云+柴本良合著，许国志协助",
+    "type": "concept"
+  },
   "十大类生产性服务业": {
     "summary": "来源：黄奇帆 2026 年演讲（笔记侠 ID: Notesman）",
     "type": "concept"
@@ -60,6 +64,14 @@
   "山西12个原始古村落": {
     "summary": "山西古村落考察实践记录"
   },
+  "工程系统工程": {
+    "summary": "来源：钱学森《论科学技术研究的组织管理与科研系统工程》（1980）第二节",
+    "type": "concept"
+  },
+  "情报信息系统工程": {
+    "summary": "来源：钱学森《论科学技术研究的组织管理与科研系统工程》（1980）第四节",
+    "type": "concept"
+  },
   "教学过程核心模型": {
     "summary": "专业教学过程的核心模型（课程设计与教学环节结构化）"
   },
@@ -69,9 +81,21 @@
   "教育研究核验提示词库": {
     "summary": "核验教育研究文献的 AI 智能体提示词模板，配套教学过程核心模型佐证研究"
   },
+  "教育系统工程": {
+    "summary": "来源：钱学森《论科学技术研究的组织管理与科研系统工程》（1980）第二节",
+    "type": "concept"
+  },
   "新范式": {
     "summary": "陆奇2024元框架：三位一体/三拐点/模型=知识/边际成本转向固定成本"
   },
+  "柴本良": {
+    "summary": "我们是其中少数几篇论文的合作者。",
+    "type": "entity"
+  },
+  "标准系统工程": {
+    "summary": "来源：钱学森《论科学技术研究的组织管理与科研系统工程》（1980）第四节",
+    "type": "concept"
+  },
   "涌现": {
     "summary": "大量简单个体在临界条件下自发诞生不可还原的全新宏观秩序"
   },
@@ -86,10 +110,22 @@
     "summary": "来源：黄奇帆 2026 年演讲",
     "type": "concept"
   },
+  "科学学": {
+    "summary": "来源：钱学森《论科学技术研究的组织管理与科研系统工程》（1980）第一节",
+    "type": "concept"
+  },
   "科技创新五阶段论": {
     "summary": "来源：黄奇帆 2026 年演讲",
     "type": "concept"
   },
+  "科研社会化": {
+    "summary": "来源：钱学森《论科学技术研究的组织管理与科研系统工程》（1980）第三节",
+    "type": "concept"
+  },
+  "科研系统工程": {
+    "summary": "来源：钱学森《论科学技术研究的组织管理与科研系统工程》（1980）",
+    "type": "concept"
+  },
   "系统工程工程实践": {
     "summary": "来源：钱学森 + 王寿云 1980s CCTV 系统工程讲座第一讲（第二节）",
     "type": "concept"
@@ -101,6 +137,14 @@
   "老红军长寿现象的精神分析": {
     "summary": "革命信念与精神状态对生命力的深层影响"
   },
+  "计量系统工程": {
+    "summary": "来源：钱学森《论科学技术研究的组织管理与科研系统工程》（1980）第四节",
+    "type": "concept"
+  },
+  "许国志": {
+    "summary": "1955 年钱学森同志在和我们中的许国志同志讨论问题时，表示了把运筹学和社会主义计划经济结合起来的想法。",
+    "type": "entity"
+  },
   "郭朝晖": {
     "summary": "技术创新管理专家，曾任职宝钢，著《我的科研经历》"
   },

tools/docs/LLM-Wiki-v2升级技术方案.md

@@ -874,6 +874,8 @@ def auto_suggest_low_confidence():
 | `.githooks/pre-commit` | P2 | Git hook |
 | `.githooks/post-merge` | P2 | Git hook |
 | `tools/data/knowledge-graph.json` | P3 | 图数据库文件 |
+| `tools/scripts/populate-relations.py` | P1+ | 批量 relations 推断（source-grouping + cross-link） |
+| `tools/scripts/enrich-citations.py` | P0+ | 批量溯源注入（citation seed + 来源节 + 存根标记） |
 
 ### 修改文件
 
@@ -882,6 +884,11 @@ def auto_suggest_low_confidence():
 | `AGENTS.md` | P0/P1 | 修复冗余 + 新增 v2 特性定义 + 更新模板 |
 | `wiki/index.md` | P0 | 更新指标数字 |
 | `wiki/log.md` | P0/P1/P2 | 追加每次操作日志 |
+| `Templates/Wiki Page Template.md` | P2+ | 纳入 Phase 1 lifecycle 字段 |
+| `Templates/People Template.md` | P2+ | 纳入 Phase 1 lifecycle 字段 |
+| `Templates/Author Template.md` | P2+ | 纳入 Phase 1 lifecycle 字段 |
+| `wiki/*.md`（375 页） | P1 | 批量填充 relations 字段 |
+| `wiki/*.md`（284 页） | P0 | 注入 `[raw:source:line]` 种子引用 |
 
 ### 依赖安装
 
@@ -890,3 +897,17 @@ pip install pyyaml  # 已安装，确认版本
 # Phase 3 可能需要：
 pip install networkx  # 轻量图分析（替代 Neo4j）
 ```
+
+---
+
+## 执行状态总览（2026-07-01）
+
+| 阶段 | 状态 | 关键指标 |
+|------|------|----------|
+| **Phase 0** — 即时修复 | ✅ 完成 | AGENTS.md 冗余清除；行号覆盖率 56%→100%；index.md 指标修正 |
+| **Phase 1** — 短期升级 | ✅ 完成 | relations 字段定义；manage-relations/graph-search/check-staleness/detect-conflicts 脚本就位 |
+| **Phase 2** — 中期自动化 | ✅ 完成 | pre-commit/post-merge hooks；validate-frontmatter/weekly-lint/promote-knowledge 脚本；结晶机制+知识提级入 AGENTS.md |
+| **Phase 3** — 长期进阶 | ✅ 完成 | knowledge-graph.py（505节点/646边）；review-pages.py（LLM+静态双模式）；graph-search.py --use-graph |
+| **P0 普及溯源** | ✅ 完成 | enrich-citations.py：引用覆盖 79→405 页（15.6%→80.2%） |
+| **P1 激活关系图** | ✅ 完成 | populate-relations.py：375 页 +644 条 relations，图 2→646 边 |
+| **P2 落地生命周期** | ✅ 完成 | 3 个模板更新，Phase 1 字段入模板规范 |

tools/scripts/enrich-citations.py

@@ -0,0 +1,207 @@
+#!/usr/bin/env python
+# -*- coding: utf-8 -*-
+"""
+enrich-citations.py — Phase 3 P0: 批量为零引用页面补溯源链接
+
+策略：
+  1. **来源节补全**：无 `## 来源` 节的页面，自动添加，包含指向 raw source 的 wikilink
+  2. **首段引用注入**：若 raw source 文件存在，提取页面标题关键词，
+     在 raw 文件中定位匹配行，注入首条 `[raw:source:行号]` 作为溯源种子
+  3. **存根页标记**：body < 50 字的页面添加 `> [!warning] 存根页面` callout
+
+安全措施：
+  - 幂等：已有 `## 来源` 节或 `[raw:` 引用则跳过
+  - --dry-run / --apply
+  - 仅处理 wiki/*.md（排除 index/log/archive）
+
+用法：
+    python tools/scripts/enrich-citations.py --dry-run
+    python tools/scripts/enrich-citations.py --apply
+"""
+import argparse
+import re
+import sys
+from pathlib import Path
+
+try:
+    import yaml
+except ImportError:
+    yaml = None
+
+VAULT = Path(r"D:\Applications\app\kepano-obsidian-main")
+WIKI = VAULT / "wiki"
+RAW = VAULT / "raw"
+EXCLUDE = {"index.md", "log.md"}
+
+
+def read_frontmatter(fp: Path) -> tuple[dict, str, str]:
+    content = fp.read_text(encoding="utf-8")
+    m = re.match(r"^---\n(.*?)\n---\n?(.*)", content, re.DOTALL)
+    if not m:
+        return {}, content, ""
+    fm_text = m.group(1)
+    body = m.group(2)
+    if yaml:
+        try:
+            front = yaml.safe_load(fm_text) or {}
+        except Exception:
+            front = {}
+    else:
+        front = {}
+    return front, body, fm_text
+
+
+def get_source_filename(front: dict) -> str:
+    """Extract raw filename from source field"""
+    source = front.get("source", "")
+    if not source:
+        return ""
+    s = str(source).strip().strip('"').strip("'")
+    s = re.sub(r"\[\[(?:raw/)?(.+?)(?:\|.*)?\]\]", r"\1", s)
+    s = re.sub(r"\.md$", "", s)
+    return s
+
+
+def find_raw_file(source_name: str) -> Path | None:
+    """Try to locate the raw file"""
+    candidates = [
+        RAW / f"{source_name}.md",
+        RAW / source_name,
+    ]
+    # Also search subdirectories
+    for pattern in [f"**/{source_name}.md", f"**/{source_name}/index.md"]:
+        matches = list(RAW.glob(pattern))
+        if matches:
+            candidates.append(matches[0])
+    for c in candidates:
+        if c.exists() and c.is_file():
+            return c
+    return None
+
+
+def find_keyword_line(raw_path: Path, page_name: str) -> str:
+    """Find a line in raw file matching the page name or key keyword"""
+    try:
+        raw_text = raw_path.read_text(encoding="utf-8")
+    except Exception:
+        return ""
+    lines = raw_text.splitlines()
+
+    # Strategy: search for page name, then aliases
+    keywords = [page_name]
+    # Also try shorter keywords
+    if len(page_name) > 4:
+        keywords.append(page_name[:4])
+
+    for kw in keywords:
+        for i, line in enumerate(lines, 1):
+            if kw in line and len(line.strip()) > 5:
+                return f"{i}"
+    return ""
+
+
+def has_source_section(body: str) -> bool:
+    return bool(re.search(r"^##\s*来源", body, re.MULTILINE))
+
+
+def has_raw_citations(body: str) -> bool:
+    return "[raw:" in body
+
+
+def is_stub(body: str) -> bool:
+    clean = re.sub(r"!\[\[.*?\]\]", "", body)
+    clean = re.sub(r"```.*?```", "", clean, flags=re.DOTALL)
+    clean = re.sub(r"\|.*?\|", "", clean)
+    clean = re.sub(r"\s+", "", clean)
+    return len(clean) < 150
+
+
+def main():
+    parser = argparse.ArgumentParser(description="Enrich wiki pages with source citations")
+    parser.add_argument("--dry-run", action="store_true", help="Preview without writing")
+    parser.add_argument("--apply", action="store_true", help="Write changes to files")
+    args = parser.parse_args()
+
+    if not args.dry_run and not args.apply:
+        args.dry_run = True
+
+    stats = {"source_section": 0, "citation_seed": 0, "stub_flag": 0, "skip_has_citation": 0, "skip_no_source": 0}
+
+    for fp in sorted(WIKI.glob("*.md")):
+        if fp.name in EXCLUDE:
+            continue
+        front, body, fm_text = read_frontmatter(fp)
+        page_name = fp.stem
+
+        if has_raw_citations(body):
+            stats["skip_has_citation"] += 1
+            continue
+
+        source_name = get_source_filename(front)
+        raw_path = find_raw_file(source_name) if source_name else None
+
+        modified = False
+        additions = []
+
+        # 1. Add citation seed if raw file exists
+        if raw_path:
+            line_num = find_keyword_line(raw_path, page_name)
+            if line_num:
+                # Find first paragraph in body and append citation
+                first_para_match = re.search(r"^(.+?)(?:\n\n|\n#)", body, re.DOTALL)
+                if first_para_match:
+                    insert_point = first_para_match.end()
+                    cite = f"[raw:{source_name}:{line_num}]"
+                    if cite not in body:
+                        body = body[:insert_point] + f" {cite}" + body[insert_point:]
+                        modified = True
+                        stats["citation_seed"] += 1
+                        additions.append(f"citation seed [raw:{source_name}:{line_num}]")
+
+        # 2. Add/fix source section
+        if not has_source_section(body):
+            source_link = front.get("source", "")
+            if source_link:
+                source_section = f"\n\n## 来源\n\n> **溯源规则**：所有数字/百分比/具体结论必须标注 `[raw:{{文件名}}:{{行号}}]` 格式。\n\n- {source_link}"
+                body = body.rstrip() + source_section
+                modified = True
+                stats["source_section"] += 1
+                additions.append("source section")
+
+        # 3. Flag stub pages
+        if is_stub(body) and not args.dry_run:
+            if "> [!warning] 存根页面" not in body:
+                stub_callout = "\n\n> [!warning] 存根页面\n> 本页内容不足，待扩充。参考同类黄金标准页面（如 [[DSpark]]、[[何长工]]）补充来源引用和正文内容。\n"
+                body = body.rstrip() + stub_callout
+                modified = True
+                stats["stub_flag"] += 1
+                additions.append("stub callout")
+
+        if not modified:
+            if not source_name:
+                stats["skip_no_source"] += 1
+            continue
+
+        if args.dry_run:
+            print(f"  [[{page_name}]]: {', '.join(additions)}")
+            continue
+
+        if args.apply:
+            # Reconstruct file
+            if yaml:
+                new_yaml = yaml.dump(front, allow_unicode=True, default_flow_style=False, sort_keys=False)
+            else:
+                new_yaml = fm_text
+            fp.write_text(f"---\n{new_yaml}---\n{body}", encoding="utf-8")
+
+    mode = "DRY RUN" if args.dry_run else "APPLIED"
+    print(f"\n=== {mode} ===")
+    print(f"  Citation seeds added: {stats['citation_seed']}")
+    print(f"  Source sections added: {stats['source_section']}")
+    print(f"  Stub flags added: {stats['stub_flag']}")
+    print(f"  Skipped (has citation): {stats['skip_has_citation']}")
+    print(f"  Skipped (no source): {stats['skip_no_source']}")
+
+
+if __name__ == "__main__":
+    main()

tools/scripts/fix-yaml-frontmatter.py

@@ -0,0 +1,199 @@
+#!/usr/bin/env python
+# -*- coding: utf-8 -*-
+"""
+fix-yaml-frontmatter.py — 批量修复 wiki frontmatter YAML 错误
+
+修复类型：
+  1. Windows 反斜杠路径 → 正斜杠（source 字段 \ → /）
+  2. 缺空格的 tag 列表项（-tag → - tag）
+  3. 断裂的 type 字段（type: place + 孤儿 - "Region" → 合并为列表）
+  4. relations 字段格式标准化（Obsidian 兼容：缩进列表 + 双引号）
+  5. frontmatter 后空行规范化
+
+用法：
+    python tools/scripts/fix-yaml-frontmatter.py --dry-run
+    python tools/scripts/fix-yaml-frontmatter.py --apply
+"""
+import argparse
+import re
+import sys
+from pathlib import Path
+
+try:
+    import yaml
+except ImportError:
+    print("ERROR: pyyaml required"); sys.exit(1)
+
+VAULT = Path(r"D:\Applications\app\kepano-obsidian-main")
+WIKI = VAULT / "wiki"
+EXCLUDE = {"index.md", "log.md"}
+
+
+class IndentedDumper(yaml.Dumper):
+    """Custom dumper that indents sequences under their key (Obsidian-friendly)"""
+    def increase_indent(self, flow=False, indentless=False):
+        return super().increase_indent(flow, False)
+
+
+def yaml_dump(front: dict) -> str:
+    """Dump frontmatter with Obsidian-compatible formatting"""
+    return yaml.dump(
+        front,
+        Dumper=IndentedDumper,
+        allow_unicode=True,
+        default_flow_style=False,
+        sort_keys=False,
+        width=1000,  # prevent line wrapping
+    )
+
+
+def split_frontmatter(content: str):
+    """Split into (frontmatter_text, body) preserving original formatting"""
+    m = re.match(r"^(---\n)(.*?)(\n---\n?)(.*)", content, re.DOTALL)
+    if not m:
+        return None, content, None
+    return m.group(2), m.group(4), m
+
+
+def fix_backslashes(fm_text: str) -> tuple[str, bool]:
+    """Fix Windows backslash paths in source field"""
+    changed = False
+    def replacer(m):
+        nonlocal changed
+        val = m.group(0)
+        if '\\' in val:
+            changed = True
+            return val.replace('\\', '/')
+        return val
+    result = re.sub(r'source:\s*".*?"', replacer, fm_text)
+    return result, changed
+
+
+def fix_dash_nospace(fm_text: str) -> tuple[str, bool]:
+    """Fix list items missing space after dash: -tag → - tag"""
+    changed = False
+    lines = fm_text.split('\n')
+    for i, line in enumerate(lines):
+        m = re.match(r'^(\s+)-(\S)', line)
+        if m:
+            lines[i] = m.group(1) + '- ' + m.group(2) + line[m.end():]
+            changed = True
+    return '\n'.join(lines), changed
+
+
+def fix_broken_type(fm_text: str) -> tuple[str, bool]:
+    """Fix orphaned type values: type: place\n- "Region" → type list"""
+    changed = False
+    lines = fm_text.split('\n')
+    result = []
+    i = 0
+    while i < len(lines):
+        line = lines[i]
+        type_match = re.match(r'^(\s*)type:\s*(\S+)', line)
+        if type_match and i + 1 < len(lines):
+            next_stripped = lines[i + 1].strip()
+            if next_stripped.startswith('- ') or next_stripped.startswith('-"'):
+                indent = type_match.group(1)
+                main_type = type_match.group(2)
+                extra = re.match(r'^-\s*"?(.+?)"?$', next_stripped)
+                extra_val = extra.group(1) if extra else next_stripped.lstrip('- ')
+                result.append(f'{indent}type:')
+                result.append(f'{indent}  - "{main_type}"')
+                result.append(f'{indent}  - "{extra_val}"')
+                changed = True
+                i += 2
+                continue
+        result.append(line)
+        i += 1
+    return '\n'.join(result), changed
+
+
+def process_file(fp: Path, apply: bool) -> dict:
+    content = fp.read_text(encoding="utf-8")
+    m = re.match(r"^(---\n)(.*?)(\n---\n?)(.*)", content, re.DOTALL)
+    if not m:
+        return {"file": fp.name, "skipped": "no frontmatter"}
+    
+    fm_text = m.group(2)
+    body = m.group(4)
+    fixes = []
+    
+    # Phase 1: Regex fixes for known error patterns
+    fm_text, changed = fix_backslashes(fm_text)
+    if changed:
+        fixes.append("backslash")
+    
+    fm_text, changed = fix_dash_nospace(fm_text)
+    if changed:
+        fixes.append("dash_nospace")
+    
+    fm_text, changed = fix_broken_type(fm_text)
+    if changed:
+        fixes.append("broken_type")
+    
+    # Phase 2: Parse → Re-dump with Obsidian-compatible formatting
+    try:
+        front = yaml.safe_load(fm_text)
+    except Exception as e:
+        return {"file": fp.name, "error": f"YAML still invalid after regex fixes: {e}"}
+    
+    if not isinstance(front, dict):
+        return {"file": fp.name, "skipped": "frontmatter not a dict"}
+    
+    # Re-dump with IndentedDumper (proper sequence indentation + double quotes)
+    new_fm = yaml_dump(front)
+    
+    # Check if re-dumping changed anything (format normalization)
+    if new_fm.rstrip() != fm_text.rstrip():
+        fixes.append("format_normalize")
+    
+    # Validate the re-dumped YAML
+    try:
+        yaml.safe_load(new_fm)
+    except Exception as e:
+        return {"file": fp.name, "error": f"Re-dumped YAML invalid: {e}"}
+    
+    if not fixes:
+        return {"file": fp.name, "skipped": "no fixes needed"}
+    
+    # Reconstruct file
+    new_content = f"---\n{new_fm}---\n\n{body.lstrip()}"
+    
+    if apply:
+        fp.write_text(new_content, encoding="utf-8")
+    
+    return {"file": fp.name, "fixes": fixes}
+
+
+def main():
+    parser = argparse.ArgumentParser(description="Fix YAML frontmatter errors in wiki")
+    parser.add_argument("--dry-run", action="store_true", help="Preview without writing")
+    parser.add_argument("--apply", action="store_true", help="Write fixes")
+    args = parser.parse_args()
+    
+    if not args.dry_run and not args.apply:
+        args.dry_run = True
+    
+    mode = "DRY RUN" if args.dry_run else "APPLYING"
+    print(f"=== {mode} ===\n")
+    
+    stats = {}
+    for fp in sorted(WIKI.glob("*.md")):
+        if fp.name in EXCLUDE:
+            continue
+        result = process_file(fp, args.apply)
+        if "fixes" in result:
+            print(f"  {result['file']}: {', '.join(result['fixes'])}")
+            for f in result["fixes"]:
+                stats[f] = stats.get(f, 0) + 1
+        elif "error" in result:
+            print(f"  ERROR: {result['file']}: {result['error']}")
+    
+    print(f"\n=== Summary ===")
+    for fix, count in sorted(stats.items()):
+        print(f"  {fix}: {count} files")
+    print(f"  Total files fixed: {sum(stats.values())}")
+
+
+if __name__ == "__main__":
+    main()

tools/scripts/graph-search.py

@@ -5,14 +5,17 @@ graph-search.py — 图感知搜索
 
 流程：
     1. qmd vsearch 获取语义匹配结果（Top 10）
-    2. 对每个结果，读取 relations 字段 → 获取相邻节点
+    2. 对每个结果，扩展相邻节点：
+       - 默认：读取各页面 relations 字段（实时，慢但零状态）
+       - --use-graph：读取预构建的 knowledge-graph.json（快，需先 build）
     3. 去重后返回（直接关联 + 关系扩展）
 
 用法：
-    python tools/scripts/graph-search.py "查询词" [--count 15] [--depth 1]
+    python tools/scripts/graph-search.py "查询词" [--count 15] [--depth 1] [--use-graph]
 
 依赖：
     - qmd 已安装（通过 node）
+    - --use-graph 需先运行 knowledge-graph.py build
 """
 import argparse
 import json
@@ -28,6 +31,7 @@ except ImportError:
 
 QMD = r'node "C:\Users\hhhh2024\AppData\Roaming\npm\node_modules\@tobilu\qmd\dist\cli\qmd.js"'
 WIKI = Path(r"D:\Applications\app\kepano-obsidian-main\wiki")
+GRAPH_DB = Path(r"D:\Applications\app\kepano-obsidian-main\tools\data\knowledge-graph.json")
 
 
 def qmd_search(query: str, count: int = 10) -> list[str]:
@@ -57,7 +61,7 @@ def get_relations(page_name: str) -> list[dict]:
 
 
 def expand_nodes(nodes: list[str], depth: int = 1) -> list[str]:
-    """从起始节点出发，沿 relations 扩展相邻节点"""
+    """从起始节点出发，沿 relations 扩展相邻节点（实时读取页面 frontmatter）"""
     expanded = list(nodes)
     frontier = list(nodes)
     for _ in range(depth):
@@ -75,11 +79,38 @@ def expand_nodes(nodes: list[str], depth: int = 1) -> list[str]:
     return expanded
 
 
+def expand_nodes_via_graph(nodes: list[str], depth: int = 1, min_conf: int = 3) -> list[str]:
+    """通过预构建的 knowledge-graph.json 扩展（更快，无 IO）"""
+    if not GRAPH_DB.exists():
+        return list(nodes)
+    graph = json.loads(GRAPH_DB.read_text(encoding="utf-8"))
+    adj: dict[str, list] = {}
+    for e in graph["edges"]:
+        adj.setdefault(e["source"], []).append((e["target"], e.get("confidence", 3)))
+        adj.setdefault(e["target"], []).append((e["source"], e.get("confidence", 3)))
+
+    expanded = list(nodes)
+    frontier = list(nodes)
+    for _ in range(depth):
+        next_frontier = []
+        for node in frontier:
+            for target, conf in adj.get(node, []):
+                if conf >= min_conf and target not in expanded:
+                    expanded.append(target)
+                    next_frontier.append(target)
+        frontier = next_frontier
+        if not frontier:
+            break
+    return expanded
+
+
 def main():
     parser = argparse.ArgumentParser(description="Graph-aware search for wiki pages")
     parser.add_argument("query", help="Search query")
     parser.add_argument("--count", type=int, default=15, help="Max results")
     parser.add_argument("--depth", type=int, default=1, help="Graph expansion depth")
+    parser.add_argument("--use-graph", action="store_true",
+                        help="Use prebuilt knowledge-graph.json instead of live frontmatter")
     args = parser.parse_args()
 
     print(f"=== Searching: {args.query} ===")
@@ -92,10 +123,20 @@ def main():
     for r in results:
         print(f"  [[{r}]]")
 
-    expanded = expand_nodes(results, depth=args.depth)
+    if args.use_graph and not GRAPH_DB.exists():
+        print(f"\n  (note: {GRAPH_DB.name} not found, run `knowledge-graph.py build`; "
+              f"falling back to live frontmatter)")
+
+    if args.use_graph and GRAPH_DB.exists():
+        expanded = expand_nodes_via_graph(results, depth=args.depth)
+        source = "knowledge-graph.json"
+    else:
+        expanded = expand_nodes(results, depth=args.depth)
+        source = "live frontmatter"
+
     new = [e for e in expanded if e not in results]
     if new:
-        print(f"\nGraph-expanded (via relations, depth={args.depth}):")
+        print(f"\nGraph-expanded (via {source}, depth={args.depth}):")
         for n in new[:args.count]:
             print(f"  [[{n}]]")
         if len(new) > args.count:

tools/scripts/knowledge-graph.py

@@ -0,0 +1,288 @@
+#!/usr/bin/env python
+# -*- coding: utf-8 -*-
+"""
+knowledge-graph.py — Phase 3 图存储层 + 图遍历搜索
+
+维护 wiki/ 的 typed relationships 图。存储为 JSON（tools/data/knowledge-graph.json），
+支持图构建、证据链反向追溯、以及融合搜索（qmd 直接匹配 + graph 关系扩展）。
+
+轻量级方案（Sqlite + JSON 混合）：当前仅用 JSON，无需外部数据库依赖。
+
+用法：
+    python tools/scripts/knowledge-graph.py build               # 从 relations 重建图
+    python tools/scripts/knowledge-graph.py stats               # 图统计
+    python tools/scripts/knowledge-graph.py evidence <页面>     # 反向追溯证据链（caused_by/supports）
+    python tools/scripts/knowledge-graph.py neighbors <页面>    # 直接邻接节点
+    python tools/scripts/knowledge-graph.py search "<查询>"     # 融合搜索（需 qmd）
+    python tools/scripts/knowledge-graph.py export-dot          # 输出 Graphviz DOT
+"""
+import argparse
+import json
+import re
+import subprocess
+import sys
+from collections import deque
+from pathlib import Path
+
+try:
+    import yaml
+except ImportError:
+    yaml = None
+
+VAULT = Path(r"D:\Applications\app\kepano-obsidian-main")
+WIKI = VAULT / "wiki"
+GRAPH_DB = VAULT / "tools" / "data" / "knowledge-graph.json"
+QMD = r'node "C:\Users\hhhh2024\AppData\Roaming\npm\node_modules\@tobilu\qmd\dist\cli\qmd.js"'
+
+EVIDENCE_EDGES = {"caused_by", "supports", "extends"}
+
+
+def read_frontmatter(fp: Path) -> dict | None:
+    content = fp.read_text(encoding="utf-8")
+    m = re.match(r"^---\n(.*?)\n---", content, re.DOTALL)
+    if not m or not yaml:
+        return None
+    try:
+        return yaml.safe_load(m.group(1)) or {}
+    except Exception:
+        return None
+
+
+def build_graph() -> dict:
+    """扫描所有 wiki 页面的 relations 字段，重建完整图"""
+    graph = {"nodes": [], "edges": [], "meta": {}}
+    seen = set()
+    edges_seen = set()
+
+    for fp in sorted(WIKI.glob("*.md")):
+        if fp.name in ("index.md", "log.md"):
+            continue
+        front = read_frontmatter(fp)
+        if front is None:
+            front = {}
+        name = fp.stem
+        if name not in seen:
+            graph["nodes"].append({
+                "id": name,
+                "type": front.get("type", "unknown"),
+                "confidence": front.get("confidence", 3),
+                "status": front.get("status", "active"),
+            })
+            seen.add(name)
+
+        for rel in front.get("relations", []) or []:
+            if not isinstance(rel, dict):
+                continue
+            target = str(rel.get("target", "")).strip("[]")
+            if not target or target == name:
+                continue
+            if target not in seen:
+                graph["nodes"].append({
+                    "id": target,
+                    "type": "unknown",
+                    "confidence": 3,
+                    "status": "unknown",
+                })
+                seen.add(target)
+            edge_key = (name, target, rel.get("type", "related"))
+            if edge_key in edges_seen:
+                continue
+            edges_seen.add(edge_key)
+            graph["edges"].append({
+                "source": name,
+                "target": target,
+                "type": rel.get("type", "related"),
+                "confidence": rel.get("confidence", 3),
+            })
+
+    graph["meta"] = {
+        "node_count": len(graph["nodes"]),
+        "edge_count": len(graph["edges"]),
+        "edge_types": {},
+    }
+    for e in graph["edges"]:
+        t = e["type"]
+        graph["meta"]["edge_types"][t] = graph["meta"]["edge_types"].get(t, 0) + 1
+
+    GRAPH_DB.parent.mkdir(parents=True, exist_ok=True)
+    GRAPH_DB.write_text(
+        json.dumps(graph, ensure_ascii=False, indent=2), encoding="utf-8"
+    )
+    return graph
+
+
+def load_graph() -> dict | None:
+    if not GRAPH_DB.exists():
+        return None
+    return json.loads(GRAPH_DB.read_text(encoding="utf-8"))
+
+
+def adjacency(graph: dict, directed: bool = True) -> dict:
+    """构建邻接表。directed=False 时加入反向边（标注 inverse_ 前缀）"""
+    adj: dict[str, list] = {}
+    for e in graph["edges"]:
+        adj.setdefault(e["source"], []).append((e["target"], e["type"], e.get("confidence", 3)))
+        if not directed:
+            adj.setdefault(e["target"], []).append(
+                (e["source"], "inverse_" + e["type"], e.get("confidence", 3))
+            )
+    return adj
+
+
+def find_evidence_chain(target_page: str, max_depth: int = 5) -> list[str]:
+    """从 target 反向追溯所有 caused_by/supports 关系的源头（BFS）"""
+    graph = load_graph()
+    if not graph:
+        return []
+    reverse_adj: dict[str, list] = {}
+    for e in graph["edges"]:
+        reverse_adj.setdefault(e["target"], []).append((e["source"], e["type"]))
+
+    chain: list[str] = []
+    visited = {target_page}
+    dq = deque([(target_page, 0)])
+    while dq:
+        node, depth = dq.popleft()
+        for src, rel in reverse_adj.get(node, []):
+            if rel not in EVIDENCE_EDGES:
+                continue
+            if src in visited:
+                continue
+            visited.add(src)
+            chain.append(("  " * (depth + 1)) + f"← [[{src}]] ({rel})")
+            if depth + 1 < max_depth:
+                dq.append((src, depth + 1))
+    return chain
+
+
+def graph_expanded_search(query: str, count: int = 15, min_conf: int = 3) -> list[dict]:
+    """融合搜索：qmd BM25/向量结果 → graph 关系扩展 → 去重重排序"""
+    cmd = f'{QMD} vsearch "{query}" -c wiki -n 10'
+    result = subprocess.run(cmd, capture_output=True, text=True, shell=True, timeout=30)
+    direct = [ln.strip() for ln in result.stdout.splitlines() if ln.strip()]
+
+    graph = load_graph()
+    if not graph:
+        return [{"page": p, "relevance": "direct"} for p in direct[:count]]
+
+    adj = adjacency(graph, directed=True)
+    expanded = list(direct)
+    for page in direct:
+        for target, rel_type, conf in adj.get(page, []):
+            if target not in expanded and conf >= min_conf:
+                expanded.append(target)
+
+    return [{"page": p, "relevance": "direct" if p in direct else "graph"} for p in expanded[:count]]
+
+
+def cmd_stats(args):
+    graph = load_graph()
+    if not graph:
+        print("Graph DB not found. Run 'build' first.")
+        return
+    meta = graph["meta"]
+    print(f"=== Knowledge Graph Stats ===")
+    print(f"  Nodes: {meta['node_count']}")
+    print(f"  Edges: {meta['edge_count']}")
+    print(f"\n  Edge type breakdown:")
+    for t, c in sorted(meta["edge_types"].items(), key=lambda x: -x[1]):
+        print(f"    {t}: {c}")
+    typed_nodes = sum(1 for n in graph["nodes"] if n["type"] != "unknown")
+    print(f"\n  Typed nodes: {typed_nodes} / {meta['node_count']}")
+
+
+def cmd_evidence(args):
+    chain = find_evidence_chain(args.page)
+    print(f"=== Evidence chain for [[{args.page}]] ===")
+    if not chain:
+        print("  No evidence edges (caused_by/supports/extends) found.")
+        return
+    for line in chain:
+        print(f"  {line}")
+    print(f"\n  Total upstream evidence: {len(chain)}")
+
+
+def cmd_neighbors(args):
+    graph = load_graph()
+    if not graph:
+        print("Graph DB not found. Run 'build' first.")
+        return
+    found = False
+    for e in graph["edges"]:
+        if e["source"] == args.page:
+            print(f"  --{e['type']}--> [[{e['target']}]] (conf={e.get('confidence', 3)})")
+            found = True
+    for e in graph["edges"]:
+        if e["target"] == args.page:
+            print(f"  <--{e['type']}-- [[{e['source']}]] (conf={e.get('confidence', 3)})")
+            found = True
+    if not found:
+        print(f"  [[{args.page}]] has no relations in the graph.")
+
+
+def cmd_search(args):
+    print(f"=== Fusion search: {args.query} ===")
+    results = graph_expanded_search(args.query, count=args.count)
+    direct = [r for r in results if r["relevance"] == "direct"]
+    graph_hits = [r for r in results if r["relevance"] == "graph"]
+    if direct:
+        print(f"\nDirect matches ({len(direct)}):")
+        for r in direct:
+            print(f"  [[{r['page']}]]")
+    if graph_hits:
+        print(f"\nGraph-expanded (via relations):")
+        for r in graph_hits:
+            print(f"  [[{r['page']}]]")
+    print(f"\nTotal: {len(results)}")
+
+
+def cmd_export_dot(args):
+    graph = load_graph()
+    if not graph:
+        print("Graph DB not found. Run 'build' first.")
+        return
+    print("digraph wiki {")
+    print('  rankdir=LR;')
+    for n in graph["nodes"]:
+        print(f'  "{n["id"]}";')
+    for e in graph["edges"]:
+        print(f'  "{e["source"]}" -> "{e["target"]}" [label="{e["type"]}"];')
+    print("}")
+
+
+def main():
+    parser = argparse.ArgumentParser(description="Phase 3 knowledge graph storage + traversal")
+    sub = parser.add_subparsers(dest="command", required=True)
+
+    sub.add_parser("build", help="Rebuild graph from relations")
+    sub.add_parser("stats", help="Print graph statistics")
+    p_ev = sub.add_parser("evidence", help="Trace evidence chain (caused_by/supports)")
+    p_ev.add_argument("page", help="Target page name")
+    p_nb = sub.add_parser("neighbors", help="Direct neighbors of a page")
+    p_nb.add_argument("page", help="Page name")
+    p_se = sub.add_parser("search", help="Fusion search (qmd + graph expansion)")
+    p_se.add_argument("query", help="Search query")
+    p_se.add_argument("--count", type=int, default=15)
+    sub.add_parser("export-dot", help="Export Graphviz DOT format")
+
+    args = parser.parse_args()
+
+    if args.command == "build":
+        graph = build_graph()
+        m = graph["meta"]
+        print(f"Graph built: {m['node_count']} nodes, {m['edge_count']} edges")
+        print(f"Saved to {GRAPH_DB}")
+    elif args.command == "stats":
+        cmd_stats(args)
+    elif args.command == "evidence":
+        cmd_evidence(args)
+    elif args.command == "neighbors":
+        cmd_neighbors(args)
+    elif args.command == "search":
+        cmd_search(args)
+    elif args.command == "export-dot":
+        cmd_export_dot(args)
+
+
+if __name__ == "__main__":
+    main()

tools/scripts/populate-relations.py

@@ -0,0 +1,291 @@
+#!/usr/bin/env python
+# -*- coding: utf-8 -*-
+"""
+populate-relations.py — Phase 3 P1: 批量填充 wiki 页面的 relations 字段
+
+策略（自动推断，无 LLM 依赖）：
+  1. **Source-grouping**：共享同一 raw source 的页面群，互为 part_of 关系
+     - 每组中被引用最多的页面成为 hub
+     - 其余页面 part_of → hub
+  2. **Type-based**：
+     - entity (人物) → example_of 对应 concept（如 毛泽东 example_of 教员的终极答案）
+     - place → part_of 上级地点（如 西安 part_of 陕西省）
+  3. **Cross-link inference**：页面正文中的 [[wikilink]] 若目标存在，
+     且源页面 type=concept，推断 extends 关系（仅取 top 2 避免噪声）
+
+安全措施：
+  - 幂等：已有 relations 不重复添加
+  - 仅处理 wiki/*.md（排除 index/log/archive）
+  - --dry-run 预览，--apply 执行
+  - 每页最多添加 5 条 relations（防爆裂）
+
+用法：
+    python tools/scripts/populate-relations.py --dry-run    # 预览
+    python tools/scripts/populate-relations.py --apply      # 执行
+    python tools/scripts/populate-relations.py --apply --max-per-page 3
+"""
+import argparse
+import json
+import re
+import sys
+from collections import defaultdict
+from pathlib import Path
+
+try:
+    import yaml
+except ImportError:
+    print("ERROR: pyyaml required"); sys.exit(1)
+
+VAULT = Path(r"D:\Applications\app\kepano-obsidian-main")
+WIKI = VAULT / "wiki"
+EXCLUDE = {"index.md", "log.md"}
+MAX_RELATIONS_DEFAULT = 5
+
+
+def read_frontmatter(fp: Path) -> tuple[dict, str]:
+    content = fp.read_text(encoding="utf-8")
+    m = re.match(r"^---\n(.*?)\n---", content, re.DOTALL)
+    if not m:
+        return {}, content
+    try:
+        front = yaml.safe_load(m.group(1)) or {}
+    except Exception:
+        return {}, content
+    return front, content
+
+
+def write_frontmatter(fp: Path, front: dict, body: str):
+    # Strip any existing frontmatter from body to avoid duplication
+    body_clean = re.sub(r"^---\n.*?\n---\n?", "", body, count=1, flags=re.DOTALL).lstrip()
+    new_yaml = yaml.dump(front, allow_unicode=True, default_flow_style=False, sort_keys=False)
+    fp.write_text(f"---\n{new_yaml}---\n\n{body_clean}", encoding="utf-8")
+
+
+def get_source_key(front: dict) -> str:
+    """Extract a normalized source key for grouping"""
+    source = front.get("source", "")
+    if not source:
+        return ""
+    s = str(source).strip().strip('"').strip("'")
+    # Remove [[ ]] and raw/ prefix
+    s = re.sub(r"\[\[(?:raw/)?(.+?)(?:\|.*)?\]\]", r"\1", s)
+    # Remove .md extension
+    s = re.sub(r"\.md$", "", s)
+    # Remove path prefixes
+    s = s.split("/")[-1]
+    return s
+
+
+def get_inbound_counts() -> dict[str, int]:
+    """Count inbound wikilinks for each wiki page"""
+    counts = defaultdict(int)
+    all_content = ""
+    for fp in VAULT.rglob("*.md"):
+        if ".git" in str(fp):
+            continue
+        try:
+            all_content += fp.read_text(encoding="utf-8") + "\n"
+        except Exception:
+            continue
+    for m in re.finditer(r"\[\[([^\]|#]+?)(?:[#\]|]|$)", all_content):
+        name = m.group(1).strip()
+        if name and not name.startswith("http"):
+            counts[name] += 1
+    return counts
+
+
+def extract_wikilinks(content: str) -> list[str]:
+    """Extract wikilink targets from page body (not frontmatter)"""
+    # Remove frontmatter
+    body = re.sub(r"^---\n.*?\n---", "", content, count=1, flags=re.DOTALL)
+    links = []
+    for m in re.finditer(r"\[\[([^\]|#]+?)(?:[#\]|])", body):
+        target = m.group(1).strip()
+        if target and not target.startswith("http") and not target.startswith("raw/"):
+            links.append(target)
+    return links
+
+
+def dedup_existing(front: dict) -> set[tuple[str, str]]:
+    """Get existing relation (type, target) pairs"""
+    existing = set()
+    for r in front.get("relations", []) or []:
+        if isinstance(r, dict):
+            t = r.get("target", "").strip("[]")
+            existing.add((r.get("type", ""), t))
+    return existing
+
+
+def main():
+    parser = argparse.ArgumentParser(description="Batch populate relations for wiki pages")
+    parser.add_argument("--dry-run", action="store_true", help="Preview without writing")
+    parser.add_argument("--apply", action="store_true", help="Write changes to files")
+    parser.add_argument("--max-per-page", type=int, default=MAX_RELATIONS_DEFAULT)
+    args = parser.parse_args()
+
+    if not args.dry_run and not args.apply:
+        args.dry_run = True  # default to dry-run for safety
+
+    # Phase 1: Read all pages
+    pages = {}
+    for fp in sorted(WIKI.glob("*.md")):
+        if fp.name in EXCLUDE:
+            continue
+        front, content = read_frontmatter(fp)
+        if not front:
+            continue
+        pages[fp.stem] = {"fp": fp, "front": front, "content": content}
+
+    print(f"Loaded {len(pages)} wiki pages")
+
+    # Phase 2: Compute inbound counts
+    inbound = get_inbound_counts()
+    print(f"Computed inbound counts for {len(inbound)} link targets")
+
+    # Phase 3: Group by source
+    source_groups = defaultdict(list)
+    for name, info in pages.items():
+        src = get_source_key(info["front"])
+        if src:
+            source_groups[src].append(name)
+
+    # Phase 4: Infer relations
+    additions = defaultdict(list)  # page -> list of {type, target, ...}
+
+    # 4a: Source-grouping → part_of hub
+    multi_page_groups = {s: g for s, g in source_groups.items() if len(g) > 1}
+    print(f"\nSource groups with >1 page: {len(multi_page_groups)}")
+    for src, group in multi_page_groups.items():
+        # Hub = most referenced page in group
+        hub = max(group, key=lambda p: inbound.get(p, 0))
+        for page in group:
+            if page == hub:
+                continue
+            additions[page].append({
+                "type": "part_of",
+                "target": f"[[{hub}]]",
+                "description": f"同源：{src}",
+                "confidence": 3,
+            })
+
+    # 4b: Single-source pages → extends the source concept if source is a wiki page
+    single_source = {s: g[0] for s, g in source_groups.items() if len(g) == 1}
+    for src, page in single_source.items():
+        info = pages.get(page)
+        if not info:
+            continue
+        page_type = info["front"].get("type", "")
+        if isinstance(page_type, list):
+            page_type = page_type[0] if page_type else ""
+        page_type = str(page_type).lower()
+
+        # If source matches a wiki page name, add extends
+        if src in pages and src != page:
+            src_type = pages[src]["front"].get("type", "")
+            if isinstance(src_type, list):
+                src_type = src_type[0] if src_type else ""
+            src_type = str(src_type).lower()
+
+            if page_type == "entity" and src_type in ("concept", "reference"):
+                additions[page].append({
+                    "type": "example_of",
+                    "target": f"[[{src}]]",
+                    "confidence": 3,
+                })
+            elif page_type == "concept" and src_type == "concept":
+                additions[page].append({
+                    "type": "extends",
+                    "target": f"[[{src}]]",
+                    "confidence": 3,
+                })
+            elif page_type in ("place",) and src_type in ("place", "concept"):
+                additions[page].append({
+                    "type": "part_of",
+                    "target": f"[[{src}]]",
+                    "confidence": 3,
+                })
+
+    # 4c: Cross-link inference for concept pages (top 2 wikilinks)
+    for name, info in pages.items():
+        page_type = info["front"].get("type", "")
+        if isinstance(page_type, list):
+            page_type = page_type[0] if page_type else ""
+        if str(page_type).lower() not in ("concept", "method"):
+            continue
+        links = extract_wikilinks(info["content"])
+        seen = set()
+        added = 0
+        for target in links:
+            if added >= 2:
+                break
+            if target == name or target in seen:
+                continue
+            if target not in pages:
+                continue
+            seen.add(target)
+            additions[name].append({
+                "type": "extends",
+                "target": f"[[{target}]]",
+                "confidence": 2,
+            })
+            added += 1
+
+    # Phase 5: Apply (single pass — dry-run prints, apply writes)
+    total_added = 0
+    pages_modified = 0
+    for name in sorted(pages.keys()):
+        pending = additions.get(name, [])
+        if not pending:
+            continue
+        info = pages[name]
+        front = info["front"]
+        existing = dedup_existing(front)
+
+        new_rels = list(front.get("relations", []) or [])
+        added = 0
+        for rel in pending:
+            key = (rel["type"], rel["target"].strip("[]"))
+            if key in existing:
+                continue
+            if len(new_rels) >= args.max_per_page:
+                break
+            new_rels.append(rel)
+            existing.add(key)
+            added += 1
+
+        if added == 0:
+            continue
+
+        front["relations"] = new_rels
+        pages_modified += 1
+        total_added += added
+
+        if args.dry_run:
+            print(f"  [[{name}]]: +{added}")
+            for rel in new_rels[-added:]:
+                print(f"    --{rel['type']}--> {rel['target']}")
+
+    mode = "DRY RUN" if args.dry_run else "APPLYING"
+    print(f"\n=== {mode} ===")
+    print(f"  Pages to modify: {pages_modified}")
+    print(f"  Relations to add: {total_added}")
+
+    if args.apply and pages_modified > 0:
+        written = 0
+        for name in sorted(pages.keys()):
+            info = pages[name]
+            front = info["front"]
+            rels = front.get("relations", [])
+            if not rels:
+                continue
+            # Only write if relations were newly populated (not pre-existing)
+            # Check by seeing if this page was in our additions map
+            if name not in additions:
+                continue
+            write_frontmatter(info["fp"], front, info["content"])
+            written += 1
+        print(f"  Written to {written} files")
+
+
+if __name__ == "__main__":
+    main()

tools/scripts/review-pages.py

@@ -0,0 +1,224 @@
+#!/usr/bin/env python
+# -*- coding: utf-8 -*-
+"""
+review-pages.py — Phase 3 wiki 页面评审器
+
+两种模式：
+  1. LLM 评审（需 Ollama 后端）：调用本地模型审核内容准确性，检测声明与 raw 来源
+     不匹配、跨页面矛盾、过时信息。
+  2. 静态评审（LLM 不可用时自动降级）：基于 frontmatter 规则与 raw 来源比对，
+     检测低置信度、缺 source、断链、过时等结构问题。无外部依赖。
+
+B-3.3 低置信度自动建议：confidence <= 2 的页面生成"需要更多证据"提示。
+B-3.1 LLM 评审：可检测语义偏差（被 AI 过度概括/扭曲的结论）。
+
+用法：
+    python tools/scripts/review-pages.py --stale              # 审查过时页面
+    python tools/scripts/review-pages.py --random 5           # 随机抽 5 页
+    python tools/scripts/review-pages.py --page "LLM-Wiki-v2" # 指定单页
+    python tools/scripts/review-pages.py --confidence-low     # 低置信度页面
+    python tools/scripts/review-pages.py --random 3 --llm     # 强制启用 LLM 评审
+    python tools/scripts/review-pages.py --json               # JSON 输出
+"""
+import argparse
+import json
+import random
+import re
+import shutil
+import subprocess
+import sys
+from datetime import date
+from pathlib import Path
+
+try:
+    import yaml
+except ImportError:
+    yaml = None
+
+VAULT = Path(r"D:\Applications\app\kepano-obsidian-main")
+WIKI = VAULT / "wiki"
+RAW = VAULT / "raw"
+TODAY = date.today()
+
+LLM_MODEL = "qwen2.5:7b"
+REVIEW_PROMPT = """请审核以下 Obsidian wiki 页面，输出 JSON（仅 JSON，无多余文字）：
+{
+  "accuracy_score": 1到5的整数,
+  "issues": ["具体问题描述"],
+  "confidence_match": true或false,
+  "suggested_updates": ["可执行的改进建议"]
+}
+重点关注：数字/百分比是否标注来源行号、结论是否被过度概括、frontmatter 是否完整。
+
+页面内容（截断至 4000 字符）：
+{content}
+"""
+
+
+def read_frontmatter(fp: Path) -> tuple[dict, str]:
+    content = fp.read_text(encoding="utf-8")
+    m = re.match(r"^---\n(.*?)\n---", content, re.DOTALL)
+    if not m or not yaml:
+        return {}, content
+    try:
+        return yaml.safe_load(m.group(1)) or {}, content
+    except Exception:
+        return {}, content
+
+
+def ollama_available() -> bool:
+    return shutil.which("ollama") is not None
+
+
+def llm_review(page_name: str, content: str) -> dict | None:
+    """调用 Ollama 审核单页，返回结构化结果"""
+    if not ollama_available():
+        return None
+    prompt = REVIEW_PROMPT.format(content=content[:4000])
+    try:
+        result = subprocess.run(
+            ["ollama", "run", LLM_MODEL, prompt],
+            capture_output=True, text=True, timeout=180, encoding="utf-8",
+        )
+        out = result.stdout.strip()
+        start, end = out.find("{"), out.rfind("}")
+        if start != -1 and end != -1:
+            return json.loads(out[start:end + 1])
+    except Exception:
+        pass
+    return None
+
+
+def static_review(fp: Path) -> dict:
+    """静态规则评审：无 LLM 依赖的结构化检查"""
+    front, content = read_frontmatter(fp)
+    issues = []
+    suggestions = []
+
+    cats = front.get("categories", []) or []
+    if not any("LLM Wiki" in str(c) for c in cats):
+        issues.append("categories 缺少 [[LLM Wiki]]")
+    tags = front.get("tags", []) or []
+    if "wiki" not in tags:
+        issues.append("tags 缺少 wiki")
+    if not front.get("type"):
+        issues.append("缺少 type 字段")
+    source = front.get("source", "")
+    if not source or str(source).strip('[]"') == "":
+        issues.append("缺少 source（wiki 层必填）")
+        suggestions.append(f"补充 source 指向 raw/ 来源，参考同类页面：{fp.stem}")
+
+    conf = front.get("confidence")
+    if conf is not None and isinstance(conf, (int, float)) and conf <= 2:
+        issues.append(f"低置信度 (confidence={conf})")
+        suggestions.append("需要更多证据：补充多源交叉验证或将 confidence 提升至 3+")
+
+    last = str(front.get("last_reviewed", ""))
+    if last:
+        try:
+            overdue = (TODAY - date.fromisoformat(last)).days
+            interval = front.get("review_interval_days", 180)
+            if overdue > interval:
+                issues.append(f"过时（上次审查 {last}，超期 {overdue - interval} 天）")
+                suggestions.append(f"复查内容并更新 last_reviewed 为 {TODAY.isoformat()}")
+        except ValueError:
+            issues.append(f"last_reviewed 格式无效：{last}")
+
+    no_line = re.findall(r"\[raw:[^:\]\[]+\]", content)
+    if no_line:
+        issues.append(f"{len(no_line)} 条 raw 引用缺少行号")
+        suggestions.append("运行 python tools/scripts/fix-raw-citations.py 自动补全行号")
+
+    score = 5 - min(4, len([i for i in issues if i.startswith(("低置信", "过时", "缺少 source"))]))
+    return {
+        "page": fp.stem,
+        "mode": "static",
+        "accuracy_score": max(1, score),
+        "issues": issues,
+        "confidence_match": (conf is None or conf >= 3),
+        "suggested_updates": suggestions,
+    }
+
+
+def select_pages(args) -> list[Path]:
+    all_pages = [p for p in sorted(WIKI.glob("*.md"))
+                 if p.name not in ("index.md", "log.md")]
+    if args.page:
+        fp = WIKI / f"{args.page}.md"
+        return [fp] if fp.exists() else []
+    if args.confidence_low:
+        return [p for p in all_pages
+                if (lambda c: c is not None and isinstance(c, (int, float)) and c <= 2)(
+                    read_frontmatter(p)[0].get("confidence"))]
+    if args.stale:
+        selected = []
+        for p in all_pages:
+            front, _ = read_frontmatter(p)
+            last = str(front.get("last_reviewed", ""))
+            interval = front.get("review_interval_days", 180)
+            if not last:
+                selected.append(p)
+                continue
+            try:
+                if (TODAY - date.fromisoformat(last)).days > interval:
+                    selected.append(p)
+            except ValueError:
+                selected.append(p)
+        return selected
+    if args.random:
+        return random.sample(all_pages, min(args.random, len(all_pages)))
+    return all_pages
+
+
+def main():
+    parser = argparse.ArgumentParser(description="Review wiki pages (LLM or static fallback)")
+    sel = parser.add_mutually_exclusive_group()
+    sel.add_argument("--stale", action="store_true", help="Review stale pages only")
+    sel.add_argument("--random", type=int, metavar="N", help="Randomly sample N pages")
+    sel.add_argument("--page", help="Review a single page")
+    sel.add_argument("--confidence-low", action="store_true", help="Review confidence<=2 pages")
+    parser.add_argument("--llm", action="store_true", help="Force LLM review (requires Ollama)")
+    parser.add_argument("--json", action="store_true", help="Output JSON")
+    args = parser.parse_args()
+
+    pages = select_pages(args)
+    if not pages:
+        print("No pages matched selection criteria.")
+        return
+
+    use_llm = args.llm or (not args.confidence_low)
+    llm_on = use_llm and ollama_available()
+    if use_llm and not llm_on:
+        print("(Ollama 不可用，降级为静态评审模式。安装 Ollama 以启用 LLM 语义审核。)\n")
+
+    reports = []
+    for fp in pages:
+        front, content = read_frontmatter(fp)
+        if llm_on:
+            llm = llm_review(fp.stem, content)
+            report = llm if llm else static_review(fp)
+            if report:
+                report["mode"] = "llm"
+        else:
+            report = static_review(fp)
+        report.setdefault("page", fp.stem)
+        reports.append(report)
+
+    if args.json:
+        print(json.dumps(reports, ensure_ascii=False, indent=2))
+        return
+
+    print(f"=== Review Report ({'LLM' if llm_on else 'static'} mode, {len(reports)} pages) ===\n")
+    needs_action = [r for r in reports if r["issues"]]
+    for r in reports:
+        flag = "OK " if not r["issues"] else "FIX"
+        print(f"[{flag}] [[{r['page']}]] — score {r['accuracy_score']}/5")
+        for issue in r["issues"]:
+            print(f"       • {issue}")
+        for sug in r["suggested_updates"]:
+            print(f"       → {sug}")
+    print(f"\nSummary: {len(needs_action)}/{len(reports)} pages need attention")
+
+
+if __name__ == "__main__":
+    main()

tools/scripts/sync_home_wiki.py

@@ -12,14 +12,23 @@ sync_home_wiki.py — home-wiki 知识页面 → kepano LLM Wiki 同步脚本
     concepts/  + entities/ + syntheses/  （排除 index.md）
     Daily/ MyNotes/ reports/ 不在知识同步范围。
 
-转换规则（home-wiki frontmatter → kepano Wiki 规范）：
-    categories: 加 [[LLM Wiki]]；entity/person 额外加 [[People]]
-    tags:       前缀 [wiki, {people|concept}]，subtype→concept/{subtype}（仅 concept），追加原 tags
-    created:    保留
-    source:     sources[0] → 文件路径取 stem 做 wikilink；URL/标识符原样；空则留空
-    type:       保留（concept/entity/synthesis）
-    aliases:    title 或 name
-    正文:       保留，移除 home-wiki 特有的 openclaw 自动段落（## Related、注释标记行）
+转换规则（home-wiki frontmatter → kepano Wiki 规范，Phase 1 v2 升级版）：
+    categories:    加 [[LLM Wiki]]；entity/person 额外加 [[People]]
+    tags:          前缀 [wiki, {people|concept}]，subtype→concept/{subtype}（仅 concept），追加原 tags
+    created:       保留
+    source:        sources[0] → 文件路径取 stem 做 wikilink；URL/标识符原样；空则留空
+    type:          保留（concept/entity/synthesis）
+    aliases:       title 或 name
+    confidence:    home-wiki 已有则沿用，否则默认 3
+    status:        默认 active
+    last_reviewed: 同步日期（YYYY-MM-DD）
+    review_interval_days: 180（技术类 90）
+    relations:     从 home-wiki `related: [[wikilink]]` 字段提取 → `[{type: extends, target: [[x]]}]`
+    正文:          保留，移除 home-wiki 特有的 openclaw 自动段落（## Related、注释标记行）
+
+YAML 输出格式（Obsidian 兼容）：
+    使用 IndentedDumper（子类）确保序列正确缩进（列表项 2 空格缩进），
+    wikilink 字符串加双引号，避免 Obsidian 解析异常。
 
 index 自动维护（--index）：
     用标记块 <!-- BEGIN/END home-wiki-sync --> 界定 index.md 的「概念页」「实体页」
@@ -32,7 +41,7 @@ index 自动维护（--index）：
     python tools/scripts/sync_home_wiki.py --dry-run        # 仅诊断差异
     python tools/scripts/sync_home_wiki.py                  # 同步（写入变化的页面）
     python tools/scripts/sync_home_wiki.py --index --log    # 一站式：同步+更新index+记日志
-    python tools/scripts/sync_home_wiki.py --lint           # 体检（孤儿/断链/source空值）
+    python tools/scripts/sync_home_wiki.py --lint           # 体检（孤儿/断链/source空值/缺失lifecycle字段）
 
 退出码：0 成功；1 源目录缺失；2 有错误。
 """
@@ -50,6 +59,24 @@ except ImportError:
     sys.stderr.write("ERROR: PyYAML 未安装，请运行 pip install pyyaml\n")
     sys.exit(2)
 
+
+class IndentedDumper(yaml.Dumper):
+    """Custom dumper that indents sequences under their key (Obsidian-friendly)"""
+    def increase_indent(self, flow=False, indentless=False):
+        return super().increase_indent(flow, False)
+
+
+def yaml_dump(data) -> str:
+    """Dump data with Obsidian-compatible formatting (indented sequences, double quotes)"""
+    return yaml.dump(
+        data,
+        Dumper=IndentedDumper,
+        allow_unicode=True,
+        default_flow_style=False,
+        sort_keys=False,
+        width=1000,
+    )
+
 # ---------------------------------------------------------------------------
 # 路径配置（两库位于不同位置，必须使用绝对路径）
 # ---------------------------------------------------------------------------
@@ -103,7 +130,7 @@ def _robust_parse_fm(fm_text):
         m = re.search(rf"^[ \t]*{re.escape(key)}[ \t]*:[ \t]*(.+?)[ \t]*$", fm_text, re.M)
         if m and m.group(1).strip() not in ("", "[]"):
             fm[key] = m.group(1).strip().strip("\"'")
-    for key in ("tags", "sources"):
+    for key in ("tags", "sources", "related"):
         vals = _extract_list_field(fm_text, key)
         if vals:
             fm[key] = vals
@@ -157,8 +184,38 @@ def _as_list(v):
     return [v]
 
 
+def _extract_related(fm_text):
+    """提取 home-wiki 的 `related: [[x]]` 字段为 relations 列表。
+
+    home-wiki 用 `related: [[wikilink]]` 记录关联页面（YAML 解析时 [[ 会失败），
+    转为 kepano Wiki 的 `relations: [{type: extends, target: "[[x]]"}]` 格式。
+    """
+    rels = []
+    # 匹配 related: 后跟 [[wikilink]]，支持多个
+    for m in re.finditer(r"^[ \t]*related[ \t]*:[ \t]*\[?\[([^\]|#]+)", fm_text, re.M):
+        tgt = m.group(1).strip()
+        if tgt:
+            rels.append({"type": "extends", "target": f"[[{tgt}]]", "confidence": 2})
+    # 也支持 related: 换行后列表
+    block = re.search(r"^[ \t]*related[ \t]*:[ \t]*\n((?:[ \t]+-\s*[^\n]+\n?)+)", fm_text, re.M)
+    if block:
+        for line in block.group(1).splitlines():
+            m = re.search(r"\[\[([^\]|#]+)", line)
+            if m:
+                rels.append({"type": "extends", "target": f"[[{m.group(1).strip()}]]", "confidence": 2})
+    return rels
+
+
 def convert_frontmatter(fm, default_name):
-    """home-wiki frontmatter → kepano Wiki frontmatter（有序）。"""
+    """home-wiki frontmatter → kepano Wiki frontmatter（有序 dict）。
+
+    Phase 1 lifecycle 字段（2026-07-01 新增）：
+      - confidence: 1-5（默认 3，home-wiki 有则沿用）
+      - status: active（默认）
+      - last_reviewed: 同步日期
+      - review_interval_days: 180（技术类 90）
+      - relations: 从 home-wiki related: 字段映射
+    """
     fm_type = str(fm.get("type", "concept")).strip()
     subtype = fm.get("subtype")
     if isinstance(subtype, list):
@@ -178,8 +235,6 @@ def convert_frontmatter(fm, default_name):
         type_tag = "concept"
 
     tags = ["wiki", type_tag]
-    # 仅 concept 类型把 subtype 转为 concept/{subtype} 标签
-    # （entity/person 用 people 标签 + [[People]] category，不加 concept/person）
     if subtype and fm_type == "concept":
         tags.append(f"concept/{subtype}")
     for t in _as_list(fm.get("tags")):
@@ -196,37 +251,41 @@ def convert_frontmatter(fm, default_name):
     else:
         src_val = ""
 
-    return [
-        ("categories", categories),
-        ("tags", tags),
-        ("created", str(fm.get("created", ""))),
-        ("source", src_val),
-        ("type", fm_type),
-        ("aliases", [str(title)]),
-    ]
+    today = date.today().isoformat()
+    confidence = fm.get("confidence", 3)
+    try:
+        confidence = int(confidence)
+    except (ValueError, TypeError):
+        confidence = 3
+    # 技术类（type=tool）默认 90 天审查周期
+    review_interval = 90 if fm_type == "tool" else 180
 
+    # 用普通 dict（Python 3.7+ 保持插入顺序），避免 OrderedDict YAML 序列化问题
+    pairs = {
+        "categories": categories,
+        "tags": tags,
+        "created": str(fm.get("created", "")),
+        "source": src_val,
+        "type": fm_type,
+        "aliases": [str(title)],
+        "confidence": confidence,
+        "status": "active",
+        "last_reviewed": today,
+        "review_interval_days": review_interval,
+    }
 
-def _yaml_quote(val):
-    """如果值包含 YAML 特殊字符（如 [[ wikilink 的方括号），加双引号。"""
-    s = str(val)
-    if not s:
-        return '""'
-    if '[' in s or ']' in s or '{' in s or '}' in s or ':' in s or '#' in s:
-        return f'"{s}"'
-    return s
+    return pairs
 
 
 def dump_frontmatter(pairs):
-    out = ["---"]
-    for k, v in pairs:
-        if isinstance(v, list):
-            out.append(f"{k}:")
-            for item in v:
-                out.append(f"  - {_yaml_quote(item)}")
-        else:
-            out.append(f"{k}: {_yaml_quote(v)}")
-    out.append("---")
-    return "\n".join(out) + "\n"
+    """输出 Obsidian 兼容的 frontmatter（使用 IndentedDumper 缩进序列）。"""
+    if isinstance(pairs, list):
+        from collections import OrderedDict
+        pairs = OrderedDict(pairs)
+    # 转为普通 dict（避免 OrderedDict 触发 python/object 标签）
+    data = dict(pairs) if not isinstance(pairs, dict) else pairs
+    yaml_str = yaml_dump(data)
+    return f"---\n{yaml_str}---\n\n"
 
 
 # ---------------------------------------------------------------------------
@@ -256,9 +315,23 @@ def build_page(src_path, stem):
     text = src_path.read_text(encoding="utf-8")
     fm, body = parse_doc(text)
     pairs = convert_frontmatter(fm, stem)
+    # 提取 relations（从原始 frontmatter 文本中）
+    m = re.match(r"^---\s*\n(.*?)\n---\s*\n?", text, re.S)
+    if m:
+        rels = _extract_related(m.group(1))
+        if rels:
+            # 去重：同 (type, target) 只保留第一个
+            seen = set()
+            unique_rels = []
+            for r in rels:
+                key = (r["type"], r["target"])
+                if key not in seen:
+                    seen.add(key)
+                    unique_rels.append(r)
+            pairs["relations"] = unique_rels[:5]  # 限制最多 5 条
     body_clean = strip_openclaw(body)
     # 确保正文与 frontmatter 间有空行
-    return dump_frontmatter(pairs) + "\n" + body_clean.lstrip("\n")
+    return dump_frontmatter(pairs) + body_clean.lstrip("\n")
 
 
 def _norm_full(text):
@@ -507,6 +580,27 @@ def lint():
         if m and not m.group(1).strip():
             empty_source.append(s)
 
+    # Phase 1 lifecycle 字段缺失检查（2026-07-01 新增）
+    LIFECYCLE_FIELDS = ("confidence", "status", "last_reviewed", "review_interval_days")
+    missing_lifecycle = []
+    for s in synced:
+        t = (DST_WIKI / f"{s}.md").read_text(encoding="utf-8")
+        for field in LIFECYCLE_FIELDS:
+            if not re.search(rf"^[ \t]*{re.escape(field)}[ \t]*:", t, re.M):
+                missing_lifecycle.append(f"{s} (缺 {field})")
+                break  # 一页只报一次
+
+    # relations 字段检查（可选，但建议）
+    no_relations = []
+    for s in synced:
+        t = (DST_WIKI / f"{s}.md").read_text(encoding="utf-8")
+        m = re.match(r"^---\s*\n(.*?)\n---\s*\n?", t, re.S)
+        if m:
+            fm_text = m.group(1)
+            if not re.search(r"^[ \t]*relations[ \t]*:[ \t]*\n[ \t]+-", fm_text, re.M) and \
+               not re.search(r"^[ \t]*relations[ \t]*:[ \t]*\[", fm_text, re.M):
+                no_relations.append(s)
+
     # 断链：同步页面引用的 [[x]] 在 wiki 是否存在（排除 raw 来源类、aliases、category）
     valid = allfiles | set(aliases_map.keys()) | set(synced)
     skip = {"People", "LLM Wiki", "wikilink"}
@@ -540,10 +634,16 @@ def lint():
     print(f"\n[未入 index] {len(not_in_index)} 个（概念/实体未在跨库章节）")
     for s in not_in_index:
         print(f"  - {s}")
+    print(f"\n[缺 lifecycle 字段] {len(missing_lifecycle)} 个（Phase 1 必填：confidence/status/last_reviewed/review_interval_days）")
+    for s in missing_lifecycle:
+        print(f"  - {s}")
+    print(f"\n[无 relations] {len(no_relations)} 个（建议补充以激活图遍历）")
+    for s in no_relations:
+        print(f"  - {s}")
 
-    issues = len(orphans) + len(empty_source) + len(not_in_index)
-    print(f"\n硬性问题（孤儿+空source+未入index）: {issues}")
-    print(f"软性问题（断链，多为待创建概念/raw来源）: {len(broken)}")
+    issues = len(orphans) + len(empty_source) + len(not_in_index) + len(missing_lifecycle)
+    print(f"\n硬性问题（孤儿+空source+未入index+缺lifecycle）: {issues}")
+    print(f"软性问题（断链/无relations，多为待创建概念/raw来源）: {len(broken) + len(no_relations)}")
     return issues
 
 

tools/scripts/validate-frontmatter.py

@@ -65,15 +65,22 @@ def validate_one(fp: Path) -> list[str]:
         errors.append("Missing 'wiki' in tags")
 
     # type must exist and be valid
+    # Convention: type can be scalar (e.g. 'concept') or list (e.g. ['entity', 'People', 'Author'])
+    # When list, the FIRST element is the canonical type; subsequent are role/subtype annotations
     ptype = front.get("type")
     if not ptype:
         errors.append("Missing type")
-    elif isinstance(ptype, str) and ptype not in VALID_TYPES:
-        errors.append(f"Invalid type: '{ptype}' (valid: {', '.join(VALID_TYPES)})")
+    elif isinstance(ptype, str):
+        if ptype not in VALID_TYPES:
+            errors.append(f"Invalid type: '{ptype}' (valid: {', '.join(VALID_TYPES)})")
     elif isinstance(ptype, list):
-        for t in ptype:
-            if t not in VALID_TYPES:
-                errors.append(f"Invalid type in list: '{t}'")
+        if not ptype:
+            errors.append("type list is empty")
+        else:
+            primary = ptype[0]
+            if primary not in VALID_TYPES:
+                errors.append(f"Invalid primary type: '{primary}' (valid: {', '.join(VALID_TYPES)})")
+            # Role/subtype annotations (2nd+ elements) are not validated against VALID_TYPES
 
     # source must exist
     if not front.get("source"):