feat(memory-management): 添加自动合并L2→L1功能

TevinClaw

9 hours ago a436d9c66ac79c43cde59dcf5ef7b4f08d3033d5

feat(memory-management): 添加自动合并L2→L1功能

- 实现4类内容筛选策略（决策/技术/教训/流程）
- 添加长度过滤（>200字）和结构化检查（###标题）
- 实现去重机制避免重复合并
- 修复原memory-merger路径问题

自动生成周报时会自动执行合并

 workspace/skills/memory-management/scripts/weekly_maintenance.py

@@ -2,7 +2,7 @@
"""
每周维护脚本
周一早上9:30执行，负责：
1. 运行memory-merger整理L2→L1
1. 自动合并L2→L1（基于内容筛选策略）
2. 检查L0大小
3. 生成周报
4. 发送报告（可选）
@@ -11,6 +11,7 @@
import os
import sys
import subprocess
import re
from datetime import datetime, timedelta
from pathlib import Path

@@ -20,28 +21,242 @@
    return Path.home() / ".openclaw" / "workspace"


def run_memory_merger() -> tuple:
    """运行memory-merger技能。"""
    workspace = get_workspace_path()
    merger_path = workspace / ".agents" / "skills" / "memory-merger"
def should_merge_to_l1(content: str) -> tuple:
    """
    判断L2内容是否应该合并到L1
    返回: (是否应该合并, 原因/类别)
    """
    content_lower = content.lower()
    
    if not merger_path.exists():
        return False, "memory-merger技能未安装"
    # 策略1: 关键词匹配 - 重要决策类
    decision_keywords = [
        "决策", "决定", "结论", "方案", "选择",
        "采用", "确定", "最终", "resolved", "解决",
        "关键", "重要", "核心", "原则"
    ]
    
    # 运行memory-merger
    # 策略2: 技术方案类
    tech_keywords = [
        "架构", "设计", "实现", "配置", "优化",
        "部署", "迁移", "升级", "重构", "方案"
    ]
    
    # 策略3: 经验教训类
    lesson_keywords = [
        "教训", "经验", "学习", "注意", "避免",
        "问题", "bug", "错误", "失败原因"
    ]
    
    # 策略4: 流程规范类
    process_keywords = [
        "流程", "规范", "规则", "约定", "标准",
        "红线", "必须", "禁止", "要求"
    ]
    
    # 检查匹配
    matched_keywords = []
    category = None
    
    for kw in decision_keywords:
        if kw in content_lower:
            matched_keywords.append(kw)
            category = "决策记录"
            break
    
    if not category:
        for kw in tech_keywords:
            if kw in content_lower:
                matched_keywords.append(kw)
                category = "技术方案"
                break
    
    if not category:
        for kw in lesson_keywords:
            if kw in content_lower:
                matched_keywords.append(kw)
                category = "经验教训"
                break
    
    if not category:
        for kw in process_keywords:
            if kw in content_lower:
                matched_keywords.append(kw)
                category = "流程规范"
                break
    
    # 策略5: 内容长度检查（太短的内容不值得合并）
    if len(content) < 200:
        return False, "内容过短"
    
    # 策略6: 必须有结构化标记（有###标题的才算正式记录）
    has_structure = bool(re.search(r'###\s+', content))
    if not has_structure:
        return False, "缺乏结构化标记"
    
    if category:
        return True, category
    
    return False, "未匹配合并策略"


def is_duplicate_in_l1(content: str, l1_file: Path) -> bool:
    """检查内容是否已在L1中存在（简单去重）"""
    if not l1_file.exists():
        return False
    
    try:
        result = subprocess.run(
            ["python", str(merger_path / "scripts" / "merge.py"), "memory-management"],
            capture_output=True,
            text=True,
            timeout=60
        )
        if result.returncode == 0:
            return True, result.stdout
        else:
            return False, result.stderr
        l1_content = l1_file.read_text(encoding='utf-8')
        # 提取内容前100字作为指纹
        content_fingerprint = content[:100].strip()
        # 检查L1中是否已有相似内容
        return content_fingerprint in l1_content
    except Exception:
        return False


def auto_merge_l2_to_l1() -> dict:
    """
    自动合并本周L2到L1
    返回合并统计信息
    """
    workspace = get_workspace_path()
    journal_dir = workspace / "memory" / "journal"
    milestones_dir = workspace / "memory" / "milestones"
    
    if not journal_dir.exists():
        return {"status": "no_journal", "merged": 0, "skipped": 0, "details": []}
    
    # 确保milestones目录存在
    milestones_dir.mkdir(parents=True, exist_ok=True)
    
    # 获取本周日期范围
    today = datetime.now()
    start_of_week = today - timedelta(days=today.weekday())
    
    # 本周的L1文件
    current_month = today.strftime("%Y-%m")
    l1_file = milestones_dir / f"{current_month}-weekly.md"
    
    merged_count = 0
    skipped_count = 0
    details = []
    
    # 遍历本周L2文件
    for l2_file in sorted(journal_dir.glob("*.md")):
        try:
            file_date = datetime.strptime(l2_file.stem, "%Y-%m-%d")
            if not (start_of_week <= file_date <= today):
                continue
            
            # 读取L2内容
            l2_content = l2_file.read_text(encoding='utf-8')
            
            # 按事件分割（按##标题分割）
            events = re.split(r'\n##\s+\[', l2_content)
            
            for event in events[1:]:  # 第一个通常是文件头
                event = "## [" + event  # 恢复标题标记
                
                # 提取事件标题
                title_match = re.search(r'##\s*\[.*?\]\s*(.+?)\n', event)
                title = title_match.group(1).strip() if title_match else "未命名事件"
                
                # 判断是否应该合并
                should_merge, reason = should_merge_to_l1(event)
                
                if not should_merge:
                    skipped_count += 1
                    details.append({
                        "date": l2_file.stem,
                        "title": title,
                        "action": "跳过",
                        "reason": reason
                    })
                    continue
                
                # 检查是否重复
                if is_duplicate_in_l1(event, l1_file):
                    skipped_count += 1
                    details.append({
                        "date": l2_file.stem,
                        "title": title,
                        "action": "跳过",
                        "reason": "L1中已存在"
                    })
                    continue
                
                # 执行合并
                try:
                    # 准备L1格式内容
                    l1_entry = f"""
## [{l2_file.stem}] {title}
**类别**: {reason}
**来源**: [L2详情](./journal/{l2_file.name})

### 摘要
{event[:500]}...

---
"""
                    
                    # 追加到L1文件
                    with open(l1_file, 'a', encoding='utf-8') as f:
                        f.write(l1_entry)
                    
                    merged_count += 1
                    details.append({
                        "date": l2_file.stem,
                        "title": title,
                        "action": "已合并",
                        "category": reason
                    })
                    
    except Exception as e:
        return False, str(e)
                    details.append({
                        "date": l2_file.stem,
                        "title": title,
                        "action": "失败",
                        "reason": str(e)
                    })
                    
        except ValueError:
            continue  # 文件名格式不正确
        except Exception as e:
            details.append({
                "date": l2_file.stem if 'l2_file' in locals() else "unknown",
                "title": "读取失败",
                "action": "错误",
                "reason": str(e)
            })
    
    return {
        "status": "success" if merged_count > 0 else "no_merge",
        "merged": merged_count,
        "skipped": skipped_count,
        "l1_file": str(l1_file) if merged_count > 0 else None,
        "details": details
    }


def run_memory_merger() -> tuple:
    """运行自动合并（替代原memory-merger调用）"""
    result = auto_merge_l2_to_l1()
    
    if result["status"] == "no_journal":
        return False, "未找到journal目录"
    
    if result["status"] == "success":
        summary = f"✅ 合并完成: {result['merged']} 条 → {result['l1_file']}\n"
        summary += f"⏭️  跳过: {result['skipped']} 条\n"
        summary += "\n详细记录:\n"
        for d in result['details'][-5:]:  # 只显示最后5条
            summary += f"  - [{d['date']}] {d['title']}: {d['action']}"
            if 'reason' in d:
                summary += f" ({d['reason']})"
            summary += "\n"
        return True, summary
    else:
        return False, f"未找到可合并内容（本周共扫描 {result['skipped']} 条，均不符合合并条件）"


def check_l0_size() -> dict: