@hupan56/wlkj 3.3.14 → 3.3.15

package/templates/qoder/scripts/domain/kg/extract/mybatis/all.py

@@ -1,79 +0,0 @@
-# -*- coding: utf-8 -*-
-"""MyBatis XML 提取协调器。
-
-扫所有 *Mapper.xml → 解析 → 建关系 → 返回 (entities, edges)。
-"""
-from __future__ import annotations
-import os
-import sys
-import glob
-import time
-
-_THIS = os.path.dirname(os.path.abspath(__file__))
-for _i in range(8):
-    _p = os.path.dirname(_THIS)
-    if os.path.isfile(os.path.join(_p, 'foundation', 'bootstrap.py')):
-        sys.path.insert(0, _p); break
-    _THIS = _p
-
-from domain.kg.extract.mybatis import mapper_parser as mp
-from domain.kg.extract.mybatis import relation_builder as rb
-
-
-def extract_file(filepath: str, repo_id: str):
-    """单文件提取 → (entities, edges)。"""
-    parsed = mp.parse_mapper_xml(filepath)
-    if not parsed.get('namespace_short'):
-        return [], []
-    return rb.build_relations(parsed, repo_id, filepath)
-
-
-def extract_repo(repo_dir: str, repo_id: str, logger=print):
-    """扫 repo 下所有 mapper XML → 提取。"""
-    t0 = time.time()
-    patterns = [
-        os.path.join(repo_dir, '**', '*Mapper.xml'),
-        os.path.join(repo_dir, '**', '*mapper*.xml'),
-        os.path.join(repo_dir, '**', 'mapper', '*.xml'),
-    ]
-    xml_files = set()
-    for pat in patterns:
-        xml_files.update(glob.glob(pat, recursive=True))
-    # 跳过编译产物 target/classes 和测试资源
-    xml_files = [f for f in xml_files if '/target/' not in f.replace('\\', '/')
-                 and '\\target\\' not in f.lower()
-                 and '/test/' not in f.replace('\\', '/').lower()]
-    total = len(xml_files)
-    logger('  [mybatis] 扫描 %d 个 mapper XML (repo=%s)' % (total, repo_id))
-
-    all_entities, all_edges = [], []
-    err = 0
-    for xf in xml_files:
-        try:
-            es, eds = extract_file(xf, repo_id)
-            all_entities += es
-            all_edges += eds
-        except Exception as e:
-            err += 1
-            if err <= 2:
-                logger('  [mybatis] 失败 %s: %s' % (xf, str(e)[:50]))
-
-    dt = time.time() - t0
-    tables = set()
-    for e in all_entities:
-        if e['type'] == 'TABLE':
-            tables.add(e['canonical'])
-    logger('  [mybatis] 完成: %d 实体 / %d 边 / %d 表 / %d 错 / %.1fs' % (
-        len(all_entities), len(all_edges), len(tables), err, dt))
-    return all_entities, all_edges
-
-
-if __name__ == '__main__':
-    import argparse, json
-    ap = argparse.ArgumentParser()
-    ap.add_argument('repo_dir')
-    ap.add_argument('--repo-id', default='')
-    args = ap.parse_args()
-    rid = args.repo_id or os.path.basename(args.repo_dir.rstrip('/\\'))
-    es, eds = extract_repo(args.repo_dir, rid)
-    print(json.dumps({'entities': len(es), 'edges': len(eds)}, ensure_ascii=False, indent=2))

package/templates/qoder/scripts/domain/kg/extract/mybatis/mapper_parser.py

@@ -1,99 +0,0 @@
-# -*- coding: utf-8 -*-
-"""mapper XML 解析器: namespace + resultMap + 方法→SQL 映射。
-
-输出每个 mapper 文件的结构化信息, 供 relation_builder 建 queries 边 + COLUMN 实体。
-"""
-from __future__ import annotations
-import os
-import re
-import xml.etree.ElementTree as ET
-
-from . import sql_extractor as _sq
-
-
-def _extract_table_names(sql):
-    """从 SQL 提取表名 (转发到 sql_extractor)。"""
-    return _sq.extract_table_names(sql)
-
-
-def parse_mapper_xml(filepath: str) -> dict:
-    """解析一个 mapper XML。
-
-    Returns:
-        {
-            'namespace': 'com.fywl.ics.demo.mapper.TestDemoMapper',
-            'namespace_short': 'TestDemoMapper',
-            'statements': [{'id':'customPageList','type':'select','sql':'SELECT * FROM test_demo...', 'tables':['test_demo']}],
-            'result_maps': {'TestDemoResult': [{'property':'id','column':'id'}, ...]},
-            'result_types': ['com.fywl.ics.demo.domain.vo.TestDemoVo'],
-        }
-    """
-    try:
-        tree = ET.parse(filepath)
-        root = tree.getroot()
-    except Exception:
-        return _empty_result()
-
-    namespace = root.get('namespace', '') or ''
-    namespace_short = namespace.rsplit('.', 1)[-1] if namespace else os.path.basename(filepath).replace('.xml', '')
-
-    statements = []
-    result_maps = {}
-    result_types = set()
-
-    for tag in ('select', 'insert', 'update', 'delete'):
-        for stmt in root.findall(tag):
-            sid = stmt.get('id', '')
-            sql_text = _extract_sql_text(stmt)
-            tables = _extract_table_names(sql_text)
-            if stmt.get('resultType'):
-                result_types.add(stmt.get('resultType'))
-            statements.append({
-                'id': sid, 'type': tag, 'sql': sql_text, 'tables': tables,
-            })
-
-    for rm in root.findall('resultMap'):
-        rm_id = rm.get('id', '')
-        cols = []
-        for child_tag in ('id', 'result'):
-            for child in rm.findall(child_tag):
-                cols.append({
-                    'property': child.get('property', ''),
-                    'column': child.get('column', ''),
-                })
-        if rm.get('type'):
-            result_types.add(rm.get('type'))
-        result_maps[rm_id] = cols
-
-    return {
-        'namespace': namespace,
-        'namespace_short': namespace_short,
-        'statements': statements,
-        'result_maps': result_maps,
-        'result_types': list(result_types),
-    }
-
-
-def _extract_sql_text(node) -> str:
-    """提取 statement 节点的完整 SQL 文本（含 <if>/<foreach> 展开）。"""
-    parts = []
-    if node.text:
-        parts.append(node.text)
-    for child in node:
-        # <if>/<foreach>/<choose>/<where>/<set>/<trim> 等, 取 text + tail
-        if child.text:
-            parts.append(child.text)
-        # 递归子节点 (foreach 里可能有 if)
-        for sub in child:
-            if sub.text:
-                parts.append(sub.text)
-            if sub.tail:
-                parts.append(sub.tail)
-        if child.tail:
-            parts.append(child.tail)
-    return ' '.join(p.strip() for p in parts if p and p.strip())
-
-
-def _empty_result():
-    return {'namespace': '', 'namespace_short': '', 'statements': [],
-            'result_maps': {}, 'result_types': []}

package/templates/qoder/scripts/domain/kg/extract/mybatis/relation_builder.py

@@ -1,69 +0,0 @@
-# -*- coding: utf-8 -*-
-"""Mapper ↔ Entity ↔ Table 三向关联构建器。
-
-把 mapper_parser 的结构化信息 + sql_extractor 的表名 →
-生成 entities (TABLE/COLUMN) + edges (queries/maps_to/returns)。
-"""
-from __future__ import annotations
-from . import mapper_parser as mp
-from . import sql_extractor as sq
-
-
-def build_relations(parsed: dict, repo_id: str, filepath: str):
-    """从单个 mapper XML 的解析结果构建实体 + 边。
-
-    Returns: (entities, edges)
-      entities: TABLE / COLUMN 实体
-      edges:    queries (Mapper→Table) / column_maps (COLUMN 关联)
-    """
-    entities, edges = [], []
-    ns_short = parsed.get('namespace_short', '')
-    ns_full = parsed.get('namespace', '')
-    if not ns_short:
-        return entities, edges
-
-    # Mapper 实体 id (对齐 GOAL 2 的 entity_id 规则: MAP:repo:Name)
-    mapper_id = 'MAP:%s:%s' % (repo_id, ns_short)
-
-    # ===== queries 边: Mapper → 每个被访问的 TABLE =====
-    all_tables = set()
-    for stmt in parsed.get('statements', []):
-        for tbl in stmt.get('tables', []):
-            all_tables.add(tbl)
-
-    for tbl in all_tables:
-        tbl_id = 'TBL:%s:%s' % (repo_id, tbl)
-        # TABLE 实体 (如果 GOAL 2 没建过, 这里补建)
-        entities.append({
-            'id': tbl_id, 'repo_id': repo_id, 'type': 'TABLE',
-            'canonical': tbl, 'cn': '',
-            'props': {'from': 'mapper_xml', 'file': filepath},
-        })
-        # queries 边
-        edges.append({
-            'from_id': mapper_id, 'to_id': tbl_id,
-            'edge_type': 'queries',
-            'from_repo': repo_id, 'to_repo': repo_id,
-            'confidence': 1.0, 'source': 'mapper_xml',
-            'props': {'via': 'mybatis_xml'},
-        })
-
-    # ===== COLUMN 实体 (从 resultMap 提取) =====
-    for rm_id, cols in parsed.get('result_maps', {}).items():
-        for col_map in cols:
-            col_name = col_map.get('column', '')
-            if not col_name or '$' in col_name:
-                continue
-            # COLUMN 实体 (关联到第一个被访问的表, 兜底)
-            hint_table = list(all_tables)[:1]
-            tbl_for_col = hint_table[0] if hint_table else 'unknown'
-            col_id = 'COL:%s:%s.%s' % (repo_id, tbl_for_col, col_name)
-            entities.append({
-                'id': col_id, 'repo_id': repo_id, 'type': 'COLUMN',
-                'canonical': col_name, 'cn': '',
-                'props': {'table': tbl_for_col,
-                          'property': col_map.get('property', ''),
-                          'from': 'result_map', 'file': filepath},
-            })
-
-    return entities, edges

package/templates/qoder/scripts/domain/kg/extract/mybatis/sql_extractor.py

@@ -1,78 +0,0 @@
-# -*- coding: utf-8 -*-
-"""SQL 文本提取器: 从 SQL 提取表名 + 列引用 + 操作类型。
-
-正则匹配 MyBatis 动态 SQL 里残留的 ${ew.customSqlSegment} 等,
-但 FROM/INTO/UPDATE/JOIN 后的表名通常稳定可提。
-"""
-from __future__ import annotations
-import re
-
-# 表名提取 (FROM/JOIN/INTO/UPDATE 后跟标识符)
-_TABLE_PATTERNS = [
-    re.compile(r'\bFROM\s+`?([a-z_][a-z0-9_]*)`?', re.IGNORECASE),
-    re.compile(r'\bJOIN\s+`?([a-z_][a-z0-9_]*)`?', re.IGNORECASE),
-    re.compile(r'\bINTO\s+`?([a-z_][a-z0-9_]*)`?', re.IGNORECASE),
-    re.compile(r'\bUPDATE\s+`?([a-z_][a-z0-9_]*)`?\s+SET', re.IGNORECASE),
-]
-
-# 排除: 显然不是表名的 (变量/占位符/SQL关键字)
-_NOT_TABLES = {'dual', 'select', 'where', 'set', 'values', 'from', 'join',
-               'and', 'or', 'not', 'in', 'as', 'on', 'left', 'right', 'inner',
-               'outer', 'group', 'order', 'limit', 'distinct', 'case', 'when',
-               'then', 'else', 'end', 'null', 'true', 'false'}
-
-
-def extract_table_names(sql: str) -> list:
-    """从 SQL 文本提取所有被访问的表名。
-
-    Returns: ['test_demo', 'sys_user'] (去重保序)
-    """
-    if not sql:
-        return []
-    found = []
-    seen = set()
-    for pat in _TABLE_PATTERNS:
-        for m in pat.finditer(sql):
-            tbl = m.group(1).lower()
-            if tbl in _NOT_TABLES or tbl.startswith('$') or tbl.startswith('#'):
-                continue
-            # 排除太短或含特殊字符的 (变量插值残留)
-            if len(tbl) < 2 or '$' in tbl or '{' in tbl:
-                continue
-            if tbl not in seen:
-                seen.add(tbl)
-                found.append(tbl)
-    return found
-
-
-def extract_column_refs(sql: str) -> list:
-    """提取 SQL 里引用的列名 (粗粒度, 主要看 SELECT 字段列表)。
-
-    Returns: ['id', 'user_id', 'name'] (去重)
-    """
-    if not sql:
-        return []
-    cols = set()
-    # SELECT field1, field2 FROM ... (字段列表)
-    m = re.search(r'SELECT\s+(DISTINCT\s+)?(.*?)\s+FROM', sql, re.IGNORECASE | re.DOTALL)
-    if m:
-        field_part = m.group(2)
-        if field_part.strip() != '*':
-            for token in re.split(r'[,\s]+', field_part):
-                token = token.strip().strip('`')
-                # 别名 as xx → 取前半
-                token = token.split(' as ')[0].split(' AS ')[0].strip()
-                if re.match(r'^[a-z_][a-z0-9_]*$', token, re.I) and len(token) >= 2:
-                    cols.add(token.lower())
-    # xx.column 形式 (表别名.列)
-    for m in re.finditer(r'\b[a-z]+\s*\.\s*([a-z_][a-z0-9_]*)', sql, re.IGNORECASE):
-        cols.add(m.group(1).lower())
-    return sorted(cols)
-
-
-def classify_operation(stmt_type: str) -> str:
-    """MyBatis statement 类型 → 操作分类。"""
-    return {
-        'select': 'READ', 'insert': 'WRITE',
-        'update': 'WRITE', 'delete': 'WRITE',
-    }.get(stmt_type.lower(), 'UNKNOWN')

package/templates/qoder/scripts/domain/kg/extract/prd/__init__.py

@@ -1,8 +0,0 @@
-# -*- coding: utf-8 -*-
-"""PRD 业务知识提取 (GOAL 4)。
-
-子模块:
-  prd_llm_extract  PRD 正文 → LLM 抽功能点/字段/规则/端点/权限
-  prd_chunk_embed  PRD 分块 + embedding (LightRAG vector 路)
-  req_anchor       REQ-ID 强关联 (PRD/commit/代码注释 → anchors 表)
-"""

package/templates/qoder/scripts/domain/kg/extract/prd/prd_chunk_embed.py

@@ -1,105 +0,0 @@
-# -*- coding: utf-8 -*-
-"""PRD 分块 + embedding → LightRAG vector 路。
-
-让 rag_search 能搜到 PRD 正文 (语义召回), 不再只搜标题。
-复用工作台 rag.py 的 embed_texts (DashScope text-embedding-v3, 1024维)。
-"""
-from __future__ import annotations
-import os
-import sys
-
-_THIS = os.path.dirname(os.path.abspath(__file__))
-for _i in range(8):
-    _p = os.path.dirname(_THIS)
-    if os.path.isfile(os.path.join(_p, 'foundation', 'bootstrap.py')):
-        sys.path.insert(0, _p); break
-    _THIS = _p
-
-
-def chunk_text(text: str, size: int = 500, overlap: int = 50) -> list:
-    """滑窗分块。Returns: [chunk1, chunk2, ...]"""
-    if not text:
-        return []
-    chunks = []
-    i = 0
-    while i < len(text):
-        chunk = text[i:i + size]
-        chunks.append(chunk)
-        i += size - overlap
-    return chunks
-
-
-def embed_prd(prd_id: str, content_md: str, repo_id: str = ''):
-    """PRD 分块 + embedding → 写 PG embeddings 表。
-
-    Returns: {chunks, embedded} 失败返回 {'chunks':0,'embedded':0}
-    """
-    chunks = chunk_text(content_md)
-    if not chunks:
-        return {'chunks': 0, 'embedded': 0}
-
-    # 复用工作台 rag.py 的 embed_texts
-    try:
-        ws_backend = os.path.join(os.getcwd(), 'wlinkj-workspace', 'backend')
-        sys.path.insert(0, ws_backend)
-        from app.rag import embed_texts
-    except Exception:
-        # 兜底: 直接调 DashScope embedding API
-        return _embed_fallback(prd_id, chunks, repo_id)
-
-    vecs = embed_texts(chunks)
-    if not vecs:
-        return {'chunks': len(chunks), 'embedded': 0}
-
-    # 写 PG embeddings 表
-    _upsert_embeddings(prd_id, chunks, vecs, repo_id)
-    return {'chunks': len(chunks), 'embedded': len(vecs)}
-
-
-def _embed_fallback(prd_id, chunks, repo_id):
-    """直接调 DashScope text-embedding-v3 (rag.py 不可用时)。"""
-    import json, urllib.request
-    key = os.environ.get('DASHSCOPE_API_KEY', '')
-    if not key:
-        return {'chunks': len(chunks), 'embedded': 0}
-    # DashScope embedding API
-    url = 'https://dashscope.aliyuncs.com/compatible-mode/v1/embeddings'
-    try:
-        req = urllib.request.Request(
-            url,
-            data=json.dumps({
-                'model': 'text-embedding-v3',
-                'input': chunks[:10],  # 单次上限10条
-                'dimensions': 1024,
-            }).encode('utf-8'),
-            headers={'Authorization': 'Bearer ' + key, 'Content-Type': 'application/json'},
-        )
-        with urllib.request.urlopen(req, timeout=30) as resp:
-            r = json.loads(resp.read())
-        vecs = [d['embedding'] for d in r['data']]
-        _upsert_embeddings(prd_id, chunks[:len(vecs)], vecs, repo_id)
-        return {'chunks': len(chunks), 'embedded': len(vecs)}
-    except Exception:
-        return {'chunks': len(chunks), 'embedded': 0}
-
-
-def _upsert_embeddings(prd_id, chunks, vecs, repo_id):
-    """写 PG embeddings 表 (pgvector)。"""
-    try:
-        from domain.kg.extract.java import pg_upsert
-        engine, _ = pg_upsert._get_pg_engine()
-        if engine is None:
-            return
-        from sqlalchemy import text
-        import json
-        with engine.begin() as conn:
-            for chunk, vec in zip(chunks, vecs):
-                vec_str = '[' + ','.join(str(v) for v in vec) + ']'
-                eid = 'PRDEMB:%s:%d' % (prd_id[:12], hash(chunk) & 0xffff)
-                conn.execute(text("""
-                    INSERT INTO embeddings (entity_id, repo_id, kind, content, vec, embedded_at)
-                    VALUES (:id, :repo, 'prd', :content, CAST(:vec AS VECTOR), now())
-                    ON CONFLICT (entity_id) DO UPDATE SET content=EXCLUDED.content, vec=EXCLUDED.vec, embedded_at=now()
-                """), {'id': eid, 'repo': repo_id, 'content': chunk[:500], 'vec': vec_str})
-    except Exception as e:
-        print('  [prd-embed] 写入跳过:', str(e)[:60])

package/templates/qoder/scripts/domain/kg/extract/prd/prd_llm_extract.py

@@ -1,153 +0,0 @@
-# -*- coding: utf-8 -*-
-"""PRD 正文 → LLM 抽取功能点/字段/规则/端点/权限。
-
-输入: PRD markdown 正文
-输出: 结构化知识 (FEATURE/CONSTRAINT/PERMISSION 实体)
-用 DashScope qwen-plus (已接入) 做结构化抽取。
-
-被 asset/prd_importer.py 调用 (灌资产中心 prds 表时, 每个PRD正文过一遍LLM)。
-"""
-from __future__ import annotations
-import os
-import sys
-import json
-import re
-import urllib.request
-
-_THIS = os.path.dirname(os.path.abspath(__file__))
-for _i in range(8):
-    _p = os.path.dirname(_THIS)
-    if os.path.isfile(os.path.join(_p, 'foundation', 'bootstrap.py')):
-        sys.path.insert(0, _p); break
-    _THIS = _p
-
-PRD_EXTRACT_PROMPT = """从以下PRD正文提取结构化知识, 只返回严格JSON(不要markdown代码块, 不要解释), 格式:
-{"features":[{"name":"资产列表分页","desc":"支持按页查看"}],"fields":[{"name":"owner_id","label":"资产归属人","rule":"必填"}],"rules":[{"rule":"金额>1万需二次审批","type":"business"}],"endpoints":[{"path":"/asset/list","method":"GET","desc":"资产列表"}],"permissions":[{"code":"asset:list","desc":"资产查看权限"}]}
-若某类提取不到就返回空数组。PRD正文:
----
-%s
----"""
-
-
-def _call_qwen(prompt: str, timeout: int = 60) -> str:
-    """调 DashScope qwen-plus, 返回文本。失败返回 ''。"""
-    key = os.environ.get('DASHSCOPE_API_KEY', '')
-    base = os.environ.get('ALI_CHAT_BASE', '').rstrip('/')
-    model = os.environ.get('ALI_CHAT_MODEL', 'qwen-plus')
-    if not key or not base:
-        return ''
-    req = urllib.request.Request(
-        base + '/chat/completions',
-        data=json.dumps({
-            'model': model,
-            'messages': [{'role': 'user', 'content': prompt}],
-            'max_tokens': 2000, 'temperature': 0.1,
-        }).encode('utf-8'),
-        headers={'Authorization': 'Bearer ' + key, 'Content-Type': 'application/json'},
-    )
-    try:
-        with urllib.request.urlopen(req, timeout=timeout) as resp:
-            r = json.loads(resp.read())
-            return r['choices'][0]['message']['content']
-    except Exception:
-        return ''
-
-
-def _parse_json_lenient(text: str) -> dict:
-    """宽松解析 LLM 返回的 JSON (可能有 markdown 包裹或前后缀)。"""
-    if not text:
-        return {}
-    # 剥 markdown 代码块
-    text = re.sub(r'^```(?:json)?\s*', '', text.strip())
-    text = re.sub(r'\s*```$', '', text.strip())
-    # 找第一个 { 到最后一个 }
-    s = text.find('{')
-    e = text.rfind('}')
-    if s >= 0 and e > s:
-        text = text[s:e + 1]
-    try:
-        return json.loads(text)
-    except Exception:
-        return {}
-
-
-def extract_prd_knowledge(content_md: str) -> dict:
-    """PRD 正文 → LLM 抽取 → 结构化 dict。
-
-    Returns: {features, fields, rules, endpoints, permissions}
-    """
-    if not content_md or len(content_md) < 50:
-        return {'features': [], 'fields': [], 'rules': [], 'endpoints': [], 'permissions': []}
-    # 截断过长正文 (避免超 token)
-    content = content_md[:6000]
-    raw = _call_qwen(PRD_EXTRACT_PROMPT % content)
-    parsed = _parse_json_lenient(raw)
-    # 兜底确保字段齐全
-    return {
-        'features': parsed.get('features', []),
-        'fields': parsed.get('fields', []),
-        'rules': parsed.get('rules', []),
-        'endpoints': parsed.get('endpoints', []),
-        'permissions': parsed.get('permissions', []),
-    }
-
-
-def to_entities_edges(extracted: dict, prd_id: str, project_id: str, repo_id: str = ''):
-    """把 LLM 抽取结果转成 entities + edges。
-
-    entities: FEATURE / CONSTRAINT / PERMISSION
-    edges: 全部锚到 prd_id (强关联, confidence=1.0)
-    """
-    entities, edges = [], []
-    base = 'FEAT:%s' % prd_id[:12] if prd_id else 'FEAT:unknown'
-
-    # 功能点 → FEATURE
-    for f in extracted.get('features', []):
-        fname = f.get('name', '').strip()
-        if not fname:
-            continue
-        fid = '%s:%s' % (base, fname[:20])
-        entities.append({
-            'id': fid, 'repo_id': repo_id, 'type': 'FEATURE',
-            'canonical': fname, 'cn': f.get('desc', ''),
-            'props': {'source': 'prd_llm', 'prd_id': prd_id},
-        })
-        edges.append(_anchor_edge(fid, prd_id, repo_id))
-
-    # 业务规则 → CONSTRAINT
-    for r in extracted.get('rules', []):
-        rule = r.get('rule', '').strip()
-        if not rule:
-            continue
-        cid = 'CONST:%s:%s' % (prd_id[:8], rule[:20])
-        entities.append({
-            'id': cid, 'repo_id': repo_id, 'type': 'CONSTRAINT',
-            'canonical': rule[:50], 'cn': r.get('type', ''),
-            'props': {'source': 'prd_llm', 'prd_id': prd_id, 'rule': rule},
-        })
-        edges.append(_anchor_edge(cid, prd_id, repo_id))
-
-    # 权限 → PERMISSION
-    for p in extracted.get('permissions', []):
-        code = p.get('code', '').strip()
-        if not code:
-            continue
-        pid = 'PERM:PRD:%s:%s' % (prd_id[:8], code[:20])
-        entities.append({
-            'id': pid, 'repo_id': repo_id, 'type': 'PERMISSION',
-            'canonical': code, 'cn': p.get('desc', ''),
-            'props': {'source': 'prd_llm', 'prd_id': prd_id},
-        })
-        edges.append(_anchor_edge(pid, prd_id, repo_id))
-
-    return entities, edges
-
-
-def _anchor_edge(from_id: str, prd_id: str, repo_id: str) -> dict:
-    """实体 → PRD 的强关联边。"""
-    return {
-        'from_id': from_id, 'to_id': 'prd:%s' % prd_id,
-        'edge_type': 'anchors', 'from_repo': repo_id, 'to_repo': repo_id,
-        'confidence': 1.0, 'source': 'explicit',
-        'props': {'via': 'prd_llm'},
-    }