weflow-cli 1.5.0

package/scripts/promote_ideas.py

@@ -0,0 +1,297 @@
+#!/usr/bin/env python3
+"""
+知识升级管道 — 002_Literature → 003_Ideas → 004_Permanent → 008_MOC
+
+用法:
+  python scripts/promote_ideas.py
+  python scripts/promote_ideas.py --api-key <key>
+"""
+
+import sys, os, re, json
+from pathlib import Path
+from collections import Counter, defaultdict
+from datetime import datetime
+
+SCRIPTS_DIR = os.path.dirname(os.path.abspath(__file__))
+PROJECT_ROOT = os.path.dirname(SCRIPTS_DIR)
+DEFAULT_VAULT = os.path.join(PROJECT_ROOT, 'output', 'wechat-vault')
+
+sys.path.insert(0, SCRIPTS_DIR)
+from _utils import call_deepseek
+
+
+def collect_metadata(vault_path: str) -> list[dict]:
+    """收集所有 002_Literature 笔记的元数据。"""
+    lit_dir = Path(vault_path) / '002_Literature' / 'WeChat'
+    articles = []
+    for date_dir in sorted(lit_dir.iterdir()):
+        if not date_dir.is_dir():
+            continue
+        for note_file in date_dir.glob('*.md'):
+            try:
+                content = note_file.read_text(encoding='utf-8')
+            except Exception:
+                continue
+            # 提取 frontmatter
+            fm = {}
+            if content.startswith('---'):
+                end = content.find('---', 3)
+                if end != -1:
+                    for line in content[3:end].split('\n'):
+                        if ':' in line:
+                            k, _, v = line.partition(':')
+                            k = k.strip()
+                            v = v.strip().strip('"').strip("'")
+                            fm[k] = v
+
+            # 提取概念
+            concept_match = re.findall(r'## 关联网络.*?##', content, re.DOTALL)
+            concepts = []
+            if concept_match:
+                concepts = re.findall(r'\[\[([^\]]+)\]\]', concept_match[0])
+
+            # 提取标签
+            tags = []
+            tag_match = re.search(r'tags:\s*\[(.+?)\]', content)
+            if tag_match:
+                tags = [t.strip() for t in tag_match.group(1).split(',')]
+
+            articles.append({
+                'title': fm.get('title', note_file.stem),
+                'source': fm.get('source', ''),
+                'date': date_dir.name,
+                'topic': fm.get('topic', ''),
+                'hasTopic': fm.get('hasTopic', ''),
+                'concepts': concepts,
+                'tags': tags,
+                'path': note_file,
+            })
+    return articles
+
+
+def generate_ideas(articles: list[dict], api_key: str, vault_path: str) -> int:
+    """AI 分析文章，提取跨领域研究想法 → 003_Ideas。"""
+    # 按主题分组，提取代表性样本
+    by_topic = defaultdict(list)
+    for a in articles:
+        topic = a['hasTopic'].replace('[[', '').replace(']]', '') or a.get('topic', '其他')
+        by_topic[topic].append(a)
+
+    ideas_dir = Path(vault_path) / '003_Ideas'
+    ideas_dir.mkdir(parents=True, exist_ok=True)
+    created = 0
+
+    for topic, arts in by_topic.items():
+        if len(arts) < 5:
+            continue
+        # 取代表性文章（高概念数的）
+        samples = sorted(arts, key=lambda a: len(a.get('concepts', [])), reverse=True)[:10]
+        titles = '\n'.join(f'- {a["title"][:60]} (来源: {a["source"]})' for a in samples[:8])
+
+        # 收集热门概念
+        all_concepts = []
+        for a in arts:
+            all_concepts.extend(a.get('concepts', []))
+        top_concepts = [c for c, _ in Counter(all_concepts).most_common(8) if len(c) > 2]
+
+        concepts_str = ', '.join(top_concepts[:8]) if top_concepts else '无'
+
+        prompt = f'''你是一个研究助理。分析以下 {topic} 主题的文章列表，从中提炼 1-2 个研究想法。
+
+文章列表（共 {len(arts)} 篇，样本 {len(samples)} 篇）：
+{titles}
+
+热门概念：{concepts_str}
+
+请为每个想法生成以下格式（200字内/想法）：
+
+## 想法1: [标题]
+- **背景**: 一句话描述
+- **核心思路**: 2-3句
+- **相关文章**: 从样本列表中引用2-3篇
+- **可行性**: 高/中/低
+
+只返回想法，不要其他内容。'''
+
+        try:
+            result = call_deepseek(prompt, api_key, max_tokens=500)
+        except Exception as e:
+            print(f'  [WARN] {topic} AI 失败: {e}')
+            continue
+
+        idea_name = f'idea-{topic}-研究想法.md'
+        idea_path = ideas_dir / idea_name
+        if idea_path.exists():
+            continue
+
+        idea_content = f'''---
+title: "{topic} 研究想法"
+created: {datetime.now().strftime('%Y-%m-%d')}
+type: idea
+hasTopic: [[{topic}]]
+tags: [idea, {topic.lower()}, research]
+inspired_by: [{len(arts)} articles]
+---
+
+# {topic} 研究想法
+
+> 基于 {len(arts)} 篇 {topic} 文章自动提炼
+
+{result}
+
+---
+
+## 相关文章
+
+'''
+        for a in arts[:15]:
+            idea_content += f'- [[{a["date"]}/{a["path"].name}|{a["title"][:50]}]]\n'
+
+        idea_content += f'''
+
+## 热门概念
+
+'''
+        for c in top_concepts[:10]:
+            idea_content += f'- [[{c}]]\n'
+
+        idea_path.write_text(idea_content, encoding='utf-8')
+        created += 1
+        print(f'  [{topic}] → {idea_name}')
+
+    return created
+
+
+def generate_moc(articles: list[dict], vault_path: str) -> int:
+    """为每个主题生成 Map of Content → 008_MOC。"""
+    by_topic = defaultdict(list)
+    for a in articles:
+        topic = a['hasTopic'].replace('[[', '').replace(']]', '') or a.get('topic', '其他')
+        by_topic[topic].append(a)
+
+    moc_dir = Path(vault_path) / '008_MOC'
+    moc_dir.mkdir(parents=True, exist_ok=True)
+    created = 0
+
+    for topic, arts in by_topic.items():
+        if len(arts) < 5:
+            continue
+
+        # 按日期分组
+        by_date = defaultdict(list)
+        for a in arts:
+            by_date[a['date']].append(a)
+
+        # 收集所有概念
+        all_concepts = []
+        for a in arts:
+            all_concepts.extend(a.get('concepts', []))
+        top_concepts = Counter(all_concepts).most_common(15)
+
+        moc_path = moc_dir / f'MOC-{topic}.md'
+        content = f'''---
+title: "{topic} — 主题导航"
+created: {datetime.now().strftime('%Y-%m-%d')}
+type: moc
+hasTopic: [[{topic}]]
+tags: [moc, {topic.lower()}]
+article_count: {len(arts)}
+---
+
+# {topic} — 主题导航地图
+
+> {len(arts)} 篇文章 · 覆盖 {len(by_date)} 天
+
+---
+
+## 📅 按日期
+
+'''
+        for date_str in sorted(by_date.keys(), reverse=True):
+            day_arts = by_date[date_str]
+            content += f'\n### {date_str} ({len(day_arts)} 篇)\n'
+            for a in day_arts[:8]:
+                content += f'- [[{a["date"]}/{a["path"].name}|{a["title"][:60]}]] — {a["source"]}\n'
+            if len(day_arts) > 8:
+                content += f'- ... 等 {len(day_arts)} 篇\n'
+
+        content += f'''
+
+## 🔥 热门概念
+
+'''
+        for c, count in top_concepts:
+            content += f'- [[{c}]] ({count} 次)\n'
+
+        content += f'''
+
+## 🔗 相关 MOC
+
+'''
+        for other in by_topic.keys():
+            if other != topic:
+                content += f'- [[MOC-{other}]]\n'
+
+        content += f'''
+
+## 📊 统计
+
+```dataview
+TABLE date, source, rating
+FROM "002_Literature"
+WHERE contains(hasTopic, "{topic}")
+SORT date DESC
+```
+'''
+
+        moc_path.write_text(content, encoding='utf-8')
+        created += 1
+        print(f'  [{topic}] → MOC-{topic}.md ({len(arts)} articles)')
+
+    return created
+
+
+def main():
+    sys.stdout.reconfigure(encoding='utf-8', errors='replace')
+    import argparse
+    parser = argparse.ArgumentParser(description='知识升级管道')
+    parser.add_argument('--vault', default=DEFAULT_VAULT, help='Vault 路径')
+    parser.add_argument('--api-key', default='', help='DeepSeek API key')
+    parser.add_argument('--skip-ai', action='store_true', help='跳过 AI 部分（仅生成 MOC）')
+    args = parser.parse_args()
+
+    api_key = args.api_key or os.environ.get('DEEPSEEK_API_KEY', '')
+
+    print('📊 收集文章元数据...')
+    articles = collect_metadata(args.vault)
+    print(f'  共 {len(articles)} 篇文章')
+
+    # 统计
+    by_topic = defaultdict(int)
+    for a in articles:
+        topic = a['hasTopic'].replace('[[', '').replace(']]', '') or a.get('topic', '其他')
+        by_topic[topic] += 1
+    print(f'  主题分布: {dict(by_topic)}')
+
+    # Step 1: MOC（不需要 AI）
+    print('\n🗺️ 生成 008_MOC 主题导航...')
+    moc_count = generate_moc(articles, args.vault)
+
+    # Step 2: Ideas（需要 AI）
+    if not args.skip_ai and api_key:
+        print('\n💡 生成 003_Ideas 研究想法...')
+        idea_count = generate_ideas(articles, api_key, args.vault)
+        print(f'\n✓ 完成: {moc_count} MOC + {idea_count} Ideas')
+    else:
+        idea_count = 0
+        print(f'\n✓ 完成: {moc_count} MOC (跳过 AI)')
+
+    if not api_key and not args.skip_ai:
+        print('  提示: 使用 --api-key 或设置 DEEPSEEK_API_KEY 以启用 AI 想法生成')
+
+    total = moc_count + idea_count
+    print(f'  共生成 {total} 个文件')
+
+
+if __name__ == '__main__':
+    main()

package/scripts/rag_chat.py

@@ -0,0 +1,191 @@
+#!/usr/bin/env python3
+"""
+RAG 智能聊天助手 — 基于语义搜索 + AI 引擎的对话式知识检索。
+
+用法:
+  # 单次提问
+  python scripts/rag_chat.py "上周和张三聊了什么"
+  python scripts/rag_chat.py "帮我总结最近 AI 动态" --top-k 8
+  python scripts/rag_chat.py "WRF 相关的讨论" --talker "项目群"
+
+  # 交互模式（多轮对话）
+  python scripts/rag_chat.py --interactive
+
+配置:
+  API key 从 ~/.weflow-cli/config.json 读取（deepseekApiKey / dashscopeApiKey）
+"""
+
+import sys, os, json
+
+sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
+from _utils import load_config, create_engine
+from semantic_search import search as semantic_search, build_index, INDEX_DIR, VECTORS_FILE
+
+PROMPT_TEMPLATE = """你是一个个人知识助手，可以访问用户的微信聊天记录和公众号文章。
+
+基于以下检索到的相关信息回答问题。如果信息不足以回答，请明确说明，不要编造。
+
+## 相关信息
+
+{context}
+
+## 用户问题
+
+{question}
+
+请用中文简洁回答，在关键信息处引用来源编号（如 [1]、[2]）。"""
+
+
+def format_context(results: list[dict]) -> str:
+    """将搜索结果格式化为 RAG context。"""
+    chunks = []
+    for i, item in enumerate(results, 1):
+        if item.get('type') == 'article':
+            chunks.append(
+                f"[{i}] 📄 公众号文章 | {item.get('date', '?')} | {item.get('topic', '')}\n"
+                f"标题: {item.get('title', '')}\n"
+                f"内容: {item.get('text', '')[:500]}"
+            )
+        elif item.get('type') == 'chat':
+            chunks.append(
+                f"[{i}] 💬 {item.get('talker', '?')} | {item.get('time', '?')}\n"
+                f"内容: {item.get('text', '')[:500]}"
+            )
+        else:
+            chunks.append(
+                f"[{i}] 内容: {item.get('text', str(item)[:500])}"
+            )
+    return '\n\n'.join(chunks) if chunks else '（未找到相关信息）'
+
+
+def build_prompt(question: str, results: list[dict], history: list[dict] = None) -> str:
+    """构建 RAG prompt。"""
+    context = format_context(results)
+
+    if history and len(history) > 0:
+        history_text = '\n'.join(
+            f"用户: {h['question']}\n助手: {h['answer']}"
+            for h in history[-6:]  # 最近 3 轮
+        )
+        context = f"## 对话历史\n\n{history_text}\n\n## 当前检索结果\n\n{context}"
+
+    return PROMPT_TEMPLATE.format(context=context, question=question)
+
+
+def query_rag(question: str, embed_key: str, chat_key: str, top_k: int = 10,
+              talker: str = None, history: list[dict] = None) -> dict:
+    """执行一次 RAG 查询。"""
+    # 1. 语义检索
+    results = semantic_search(question, embed_key, top_k=top_k)
+
+    # 2. 过滤 talker（如果指定）
+    if talker:
+        results = [r for r in results if talker.lower() in (r.get('talker', '') + r.get('title', '')).lower()]
+        if not results:
+            return {
+                'answer': f'未找到与 "{talker}" 相关的对话或文章。',
+                'sources': [],
+            }
+
+    # 3. 构建 prompt
+    prompt = build_prompt(question, results, history)
+
+    # 4. 调用 AI
+    config = load_config()
+    engine_type = config.get('aiEngine', 'deepseek')
+    engine = create_engine(engine_type, chat_key)
+    answer = engine.chat(prompt, max_tokens=1500)
+
+    # 5. 格式化来源
+    sources = []
+    for i, item in enumerate(results[:5], 1):
+        if item.get('type') == 'chat':
+            sources.append(f"[{i}] {item.get('talker', '?')} ({item.get('time', '?')})")
+        else:
+            sources.append(f"[{i}] {item.get('title', item.get('id', '?'))[:60]}")
+
+    return {'answer': answer, 'sources': sources}
+
+
+def interactive_mode(embed_key: str, chat_key: str):
+    """交互式多轮对话。"""
+    config = load_config()
+    engine_type = config.get('aiEngine', 'deepseek')
+
+    print(f'\n🤖 RAG 智能助手 ({engine_type})')
+    print('  输入问题开始对话，输入 /help 查看帮助，输入 /exit 退出\n')
+
+    history: list[dict] = []
+
+    while True:
+        try:
+            question = input('🔍 你: ').strip()
+        except (EOFError, KeyboardInterrupt):
+            print('\n👋 再见！')
+            break
+
+        if not question:
+            continue
+
+        if question == '/exit':
+            print('👋 再见！')
+            break
+        if question == '/help':
+            print('命令: /exit 退出, /clear 清除历史, /rebuild 重建索引\n')
+            continue
+        if question == '/clear':
+            history = []
+            print('✓ 对话历史已清除\n')
+            continue
+        if question == '/rebuild':
+            print('重建索引中...')
+            result = build_index(embed_key, full=False)
+            print(f'✓ {result}\n')
+            continue
+
+        result = query_rag(question, embed_key, chat_key, top_k=10, history=history)
+        history.append({'question': question, 'answer': result['answer']})
+
+        print(f'\n🤖 助手:\n{result["answer"]}')
+        if result.get('sources'):
+            print(f"\n📎 来源: {' | '.join(result['sources'])}")
+        print()
+
+
+def main():
+    sys.stdout.reconfigure(encoding='utf-8', errors='replace')
+    import argparse
+    parser = argparse.ArgumentParser(description='RAG 智能聊天助手')
+    parser.add_argument('question', nargs='?', help='要问的问题')
+    parser.add_argument('--interactive', '-i', action='store_true', help='交互模式')
+    parser.add_argument('--top-k', type=int, default=10, help='检索数量（默认 10）')
+    parser.add_argument('--talker', help='限定联系人/群聊')
+    parser.add_argument('--api-key', help='DeepSeek API key（优先从 config 读取）')
+    parser.add_argument('--json', action='store_true', help='JSON 输出')
+    args = parser.parse_args()
+
+    config = load_config()
+    embed_key = os.environ.get('DASHSCOPE_API_KEY', '') or config.get('dashscopeApiKey', '')
+    chat_key = args.api_key or os.environ.get('DEEPSEEK_API_KEY', '') or config.get('deepseekApiKey', '')
+
+    if not embed_key:
+        print('[ERROR] 缺少 Embedding API key。请在 ~/.weflow-cli/config.json 中设置 dashscopeApiKey')
+        sys.exit(1)
+    if not chat_key:
+        print('[ERROR] 缺少 AI API key。请在 ~/.weflow-cli/config.json 中设置 deepseekApiKey')
+        sys.exit(1)
+
+    if args.interactive or not args.question:
+        interactive_mode(embed_key, chat_key)
+    else:
+        result = query_rag(args.question, embed_key, chat_key, top_k=args.top_k, talker=args.talker)
+        if args.json:
+            print(json.dumps(result, ensure_ascii=False, indent=2))
+        else:
+            print(f'\n{result["answer"]}')
+            if result.get('sources'):
+                print(f"\n📎 来源: {' | '.join(result['sources'])}")
+
+
+if __name__ == '__main__':
+    main()