feat(templates): integrate xMemory 4-layer architecture into skill and workflow

Update templates to support xMemory hierarchical memory:

- AGENTS.md: Add themes/ directory and L3 layer to architecture diagram
- workflow.md v3.0: Add [10] 主题检索 menu option for theme search
- steps/themes.md (NEW): Theme search step with omp --level theme
- steps/search.md: Add xMemory layer selection (theme/semantic/all)
- SKILL.md v3.0: Add L3 theme assimilation flow and themes/ structure

xMemory integration points:
- Theme search: omp search 'query' --level theme [--no-expand]
- Auto-assimilation: New semantics automatically attach to themes
- Configurable thresholds in _omp/memory/themes/index.yaml

Author: Alenryuichi

cli/templates/shared/_omp/AGENTS.md

@@ -8,21 +8,37 @@ This file provides complete instructions for AI agents to use the dual-layer mem
 - View project memory: `cat _omp/memory/*.md`
 - Check workflows: `ls _omp/workflows/memory/`
 
-## Memory Architecture
+## Memory Architecture (xMemory 4-Layer)
 
 ```
-├── _omp/memory/           ← Project-level memory (this repo)
-│   ├── projectbrief.md    ← Project overview & goals
-│   ├── productContext.md  ← Product requirements & features
-│   ├── techContext.md     ← Technical stack & architecture
-│   ├── activeContext.md   ← Current work session context
-│   ├── systemPatterns.md  ← Patterns & conventions
-│   ├── decisions.yaml     ← Architecture decisions log
-│   └── progress.md        ← Task progress tracking
+├── _omp/memory/             ← Project-level memory (this repo)
+│   ├── projectbrief.md      ← Project overview & goals
+│   ├── productContext.md    ← Product requirements & features
+│   ├── techContext.md       ← Technical stack & architecture
+│   ├── activeContext.md     ← Current work session context
+│   ├── systemPatterns.md    ← Patterns & conventions
+│   ├── decisions.yaml       ← Architecture decisions log
+│   ├── progress.md          ← Task progress tracking
+│   ├── sessions/            ← L1 Episode memories
+│   └── themes/              ← L3 Theme layer (NEW)
+│       ├── index.yaml       ← Theme index & config
+│       └── embeddings.json  ← Theme centroids
 │
-└── OpenMemory MCP         ← User-level memory (personal prefs)
+├── OpenMemory MCP           ← L2 Semantic layer (Qdrant)
+│   └── User preferences, skills, cross-project memories
+│
+└── L0 Message Layer         ← Raw conversation (not persisted)
 ```
 
+### xMemory Layer Summary
+
+| Layer | Storage | Purpose |
+|-------|---------|---------|
+| L3 Theme | `themes/` | Aggregated topics from semantics |
+| L2 Semantic | OpenMemory MCP | Searchable knowledge fragments |
+| L1 Episode | `sessions/` | Conversation context |
+| L0 Message | In-memory | Raw messages (transient) |
+
 ## Trigger Signals for Auto-Extraction
 
 Extract and save when you detect:

cli/templates/shared/_omp/skills/memory-extraction/SKILL.md

@@ -1,9 +1,9 @@
 ---
 name: memory-extraction
-description: Agent-only workflow for extracting key information from conversations, code changes, and deployments into structured memory files. Automatically updates `_omp/memory/` directory and notifies other agents. Triggered automatically at conversation end or when valuable information is detected.
+description: Agent-only workflow for extracting key information from conversations, code changes, and deployments into structured memory files. Supports xMemory 4-layer architecture with theme assimilation. Automatically updates `_omp/memory/` directory and notifies other agents.
 metadata:
   author: Wendy (Workflow Builder)
-  version: "2.1"
+  version: "3.0"
   language: zh-CN
   audience: agent-only
 ---
@@ -34,18 +34,30 @@ _omp/                           # OpenMemory Plus 核心目录
 │       ├── SKILL.md
 │       ├── scripts/validate.sh
 │       └── templates/*.tmpl
-└── memory/                    # 项目级记忆
+└── memory/                     # 项目级记忆
     ├── project.yaml            # 项目配置 (SSOT)
     ├── decisions.yaml          # 技术决策记录
     ├── changelog.yaml          # 变更历史
-    └── sessions/               # 会话记录
+    ├── sessions/               # 会话记录 (L1 Episode)
+    └── themes/                 # 主题层 (L3 Theme) ← NEW
+        ├── index.yaml          # 主题索引和配置
+        └── embeddings.json     # 主题质心向量
 ```
 
-## 双层记忆架构
+## xMemory 4-Layer Architecture
+
+| 层级 | 存储位置 | 用途 |
+|------|----------|------|
+| **L3 Theme** | `_omp/memory/themes/` | 主题聚合，多个语义记忆的抽象 |
+| **L2 Semantic** | openmemory MCP (Qdrant) | 语义记忆，可检索的知识片段 |
+| **L1 Episode** | `_omp/memory/sessions/` | 情节记忆，会话上下文 |
+| **L0 Message** | 对话历史 | 原始消息（不持久化） |
+
+## 双层记忆架构（简化视图）
 
 | 系统 | 存储位置 | 用途 |
 |------|----------|------|
-| 项目级 | `_omp/memory/` | 项目配置、技术决策、变更记录 |
+| 项目级 | `_omp/memory/` | 项目配置、技术决策、变更记录、主题 |
 | 用户级 | `openmemory` MCP | 用户偏好、技能、跨项目上下文 |
 
 ## 分类规则
@@ -110,14 +122,30 @@ _omp/                           # OpenMemory Plus 核心目录
 
 ---
 
-## 🔄 自动化流程
+## 🔄 自动化流程 (xMemory Enhanced)
 
 ```
-对话/操作 → 信息检测 → 分类 → 提取 → 存储 → 通知
-     ↓           ↓        ↓       ↓       ↓       ↓
-  输入源      触发判断   类型识别  结构化  写入YAML  更新Agent
+对话/操作 → 信息检测 → 分类 → 提取 → 存储 → 主题吸附 → 通知
+     ↓           ↓        ↓       ↓       ↓        ↓          ↓
+  输入源      触发判断   类型识别  结构化  L2语义  L3主题聚合  更新Agent
 ```
 
+### xMemory 主题吸附 (L3 Assimilation)
+
+当新的语义记忆存入 L2 层后，自动触发主题吸附：
+
+```bash
+# 主题吸附流程
+1. 计算新记忆的 embedding
+2. 与现有主题 centroids 计算余弦相似度
+3. 相似度 >= 0.62 (attachThreshold): 吸附到现有主题
+4. 相似度 < 0.62: 创建新主题
+5. 主题语义数 > 10: 触发主题分裂
+6. 两主题相似度 > 0.78: 触发主题合并
+```
+
+**配置位置**: `_omp/memory/themes/index.yaml`
+
 ### Phase 1: 信息检测
 
 **检测规则**（按优先级）：
@@ -470,6 +498,7 @@ pending_memories:
 
 | 版本 | 变更 |
 |------|------|
+| v3.0 | xMemory 4 层架构：添加 L3 主题层、主题吸附流程、themes/ 目录 |
 | v2.1 | 目录重构：_omp/ 统一目录，移除 rules/，分类规则内嵌 |
 | v2.0 | 双层记忆架构：整合 openmemory MCP，智能分类路由 |
 | v1.1 | 添加错误处理、Schema 验证、模板文件、触发条件详解 |

cli/templates/shared/_omp/workflows/memory/steps/search.md

@@ -1,6 +1,6 @@
 ---
 name: search
-description: 在项目级和用户级记忆中进行语义搜索
+description: 在项目级和用户级记忆中进行语义搜索（支持 xMemory 多层架构）
 ---
 
 # Step: 搜索记忆
@@ -9,25 +9,51 @@ description: 在项目级和用户级记忆中进行语义搜索
 
 - ✅ Get search query from user if not provided
 - ✅ Search both project and user memory
+- ✅ Support xMemory layer selection (theme/semantic/all)
 - ✅ Display results with relevance scores
 
 ---
 
+## xMemory Layers
+
+| 层级 | CLI 选项 | 描述 |
+|------|----------|------|
+| L3 Theme | `--level theme` | 搜索主题层（聚合） |
+| L2 Semantic | `--level semantic` | 搜索语义层（默认） |
+| All | `--level all` | 自适应检索（L3→L2→L1） |
+
+---
+
 ## EXECUTION
 
 ### 1. Get Search Query
 
 If user provided query in their command, use it directly.
 Otherwise ask:
-> "请输入搜索关键词："
+> "请输入搜索关键词（可选：指定层级 theme/semantic/all）："
+
+### 2. Determine Search Level
 
-### 2. Search User Memory
+根据用户输入或显式参数确定搜索层级：
+- 默认: `semantic` (L2 语义层)
+- 用户说 "主题/theme": `theme` (L3 主题层)
+- 用户说 "全部/all": `all` (自适应)
 
-Call `search_memory_openmemory` with the query:
-- Get matching memories with scores
-- Sort by relevance
+### 3. Search User Memory
+
+**L2 Semantic (默认):**
+```bash
+# 使用 openmemory MCP
+search_memory_openmemory(query="{query}")
+```
+
+**L3 Theme (xMemory):**
+```bash
+omp search "{query}" --level theme
+# 不展开: omp search "{query}" --level theme --no-expand
+```
 
-### 3. Search Project Memory
+### 4. Search Project Memory
 
 Search `_omp/memory/*.md` and `_omp/memory/*.yaml` files:
 - Grep for query keywords

cli/templates/shared/_omp/workflows/memory/steps/themes.md

@@ -0,0 +1,112 @@
+---
+name: themes
+description: xMemory L3 主题层检索和管理
+---
+
+# Step: 主题检索 (xMemory L3)
+
+## EXECUTION RULES
+
+- ✅ 使用 `omp search --level theme` 搜索主题层
+- ✅ 支持 `--no-expand` 选项仅返回主题，不展开到语义层
+- ✅ 显示主题的语义聚合信息
+
+---
+
+## xMemory 4-Layer Architecture
+
+```
+L3 (Theme)    ← 本操作的目标层
+    ↓ 包含
+L2 (Semantic) ← 语义记忆（Qdrant 向量）
+    ↓ 来源
+L1 (Episode)  ← 情节记忆（会话）
+    ↓ 包含
+L0 (Message)  ← 原始消息
+```
+
+---
+
+## EXECUTION
+
+### 1. Get Search Query
+
+如果用户没有提供查询词：
+> "请输入主题搜索关键词："
+
+### 2. Search Themes
+
+使用 CLI 搜索主题层：
+
+```bash
+# 搜索主题（自动展开到语义层）
+omp search "关键词" --level theme
+
+# 仅返回主题，不展开
+omp search "关键词" --level theme --no-expand
+```
+
+### 3. Display Results
+
+```
+🎯 主题检索结果: "{query}"
+
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+
+📁 L3 主题层:
+{foreach theme}
+[{n}] 主题: {theme_label}
+     相似度: {score}
+     包含语义记忆: {semantic_count} 条
+     质心向量: [{centroid_preview}...]
+{/foreach}
+
+📚 L2 语义层 (展开):
+{foreach semantic}
+[{n}] {content_preview}
+     来源主题: {parent_theme}
+     重要度: {importance}
+{/foreach}
+
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+
+共找到 {theme_count} 个主题，{semantic_count} 条语义记忆
+```
+
+### 4. Theme Management Actions
+
+用户可继续操作：
+- `"展开 N"` → 查看主题 N 下的所有语义记忆
+- `"详情 N"` → 显示主题完整信息
+- `"搜索语义"` → 切换到 L2 语义层搜索
+
+### 5. Theme Configuration
+
+主题配置位于 `_omp/memory/themes/index.yaml`:
+
+```yaml
+# 主题配置
+attachThreshold: 0.62    # 语义吸附到主题的阈值
+mergeThreshold: 0.78     # 主题合并阈值
+expandThreshold: 0.75    # 搜索时展开到语义层的阈值
+maxThemes: 50            # 最大主题数量
+splitThreshold: 10       # 单主题语义数超过此值时分裂
+```
+
+---
+
+## FALLBACK (CLI Unavailable)
+
+如果 `omp` CLI 不可用：
+
+1. 显示提示: `"⚠️ omp CLI 不可用，请先安装: npm install -g openmemory-plus"`
+2. 尝试直接读取 `_omp/memory/themes/index.yaml`
+3. 显示原始主题数据
+
+---
+
+## RETURN TO MENU
+
+完成后提示:
+> "还需要其他操作吗？输入 **M** 返回菜单，或直接输入下一个操作"
+

cli/templates/shared/_omp/workflows/memory/workflow.md

@@ -1,7 +1,7 @@
 ---
 name: memory
-description: 记忆管理工作流 - 统一管理项目级和用户级记忆
-version: "2.0"
+description: 记忆管理工作流 - 统一管理项目级和用户级记忆（含 xMemory 4 层架构）
+version: "3.0"
 ---
 
 # Memory Management Workflow
@@ -31,14 +31,19 @@ This uses **micro-file architecture** with **menu-driven routing**:
 - `memory_folder` = `{project-root}/_omp/memory`
 - `steps_path` = `{installed_path}/steps`
 
-### ⚠️ DUAL-LAYER STORAGE RULE
+### ⚠️ xMemory 4-LAYER ARCHITECTURE
 
-**CRITICAL: This system has TWO storage layers. You MUST use BOTH appropriately!**
+**CRITICAL: This system uses xMemory 4-layer hierarchical memory architecture!**
 
 | Layer | Storage | What to Store |
 |-------|---------|---------------|
-| **项目级** | `_omp/memory/*.md` files | Project config, tech stack, decisions, URLs, paths |
-| **用户级** | openmemory MCP | User preferences, skills, cross-project habits |
+| **L3 Theme** | `_omp/memory/themes/` | 聚合主题，多个语义记忆的抽象 |
+| **L2 Semantic** | openmemory MCP (Qdrant) | 语义记忆，可检索的知识片段 |
+| **L1 Episode** | `_omp/memory/sessions/` | 情节记忆，会话上下文 |
+| **L0 Message** | 对话历史 | 原始消息（不持久化） |
+
+**项目级文件** (`_omp/memory/*.md`): 项目配置、技术栈、决策、URL、路径
+**用户级记忆** (openmemory MCP): 用户偏好、技能、跨项目习惯
 
 **DO NOT store everything to openmemory! Project-specific info MUST go to `_omp/memory/` files!**
 
@@ -95,6 +100,7 @@ This uses **micro-file architecture** with **menu-driven routing**:
 [7] 🔗 知识图谱    查看实体关系
 [8] 📦 记忆整合    合并碎片化记忆
 [9] 📊 质量指标    查看记忆健康度和指标
+[10] 🎯 主题检索   搜索 L3 主题层（xMemory）
 
 [M] 返回菜单  [X] 退出
 
@@ -122,6 +128,7 @@ Based on user input, load the corresponding step file:
 | `7` | 图谱, 关系, graph, relation | `./steps/graph.md` |
 | `8` | 整合, 合并, consolidate, merge | `./steps/consolidate.md` |
 | `9` | 指标, 质量, metrics, health | `./steps/metrics.md` |
+| `10` | 主题, theme, L3, xmemory | `./steps/themes.md` |
 | `M` | 菜单, menu | Re-display menu |
 | `X` | 退出, exit, quit | Exit workflow |