docs(blog): fix mkdocs strict nav and links

Author: rexleimo

docs-site/blog/agent-memory-design.md

@@ -165,6 +165,125 @@ def retrieve_context(query: str, max_tokens: int = 8000) -> str:
     return truncate_to_token_limit(context, max_tokens)
 ```
 
+## 完整实现示例
+
+```python
+# memory_system.py
+from dataclasses import dataclass
+from enum import Enum
+from typing import Optional
+import sqlite3
+import time
+import pickle
+
+class MemoryType(Enum):
+    WORKING = "working"
+    SESSION = "session"
+    PERSISTENT = "persistent"
+
+@dataclass
+class Memory:
+    id: Optional[int]
+    memory_type: MemoryType
+    content: str
+    embedding: list[float]
+    created_at: float
+    workspace_id: str
+
+class MemorySystem:
+    def __init__(self, db_path: str):
+        self.conn = sqlite3.connect(db_path)
+        self._init_db()
+
+    def _init_db(self):
+        self.conn.execute("""
+            CREATE TABLE IF NOT EXISTS memories (
+                id INTEGER PRIMARY KEY,
+                memory_type TEXT NOT NULL,
+                content TEXT NOT NULL,
+                embedding BLOB,
+                created_at REAL NOT NULL,
+                workspace_id TEXT NOT NULL
+            )
+        """)
+        self.conn.commit()
+
+    def save_memory(
+        self,
+        memory_type: MemoryType,
+        content: str,
+        embedding: list[float],
+        workspace_id: str
+    ) -> int:
+        """保存记忆"""
+        cursor = self.conn.execute("""
+            INSERT INTO memories (memory_type, content, embedding, created_at, workspace_id)
+            VALUES (?, ?, ?, ?, ?)
+        """, (
+            memory_type.value,
+            content,
+            pickle.dumps(embedding),
+            time.time(),
+            workspace_id
+        ))
+        self.conn.commit()
+        return cursor.lastrowid
+
+    def retrieve(
+        self,
+        query_embedding: list[float],
+        workspace_id: str,
+        memory_types: list[MemoryType] = None,
+        top_k: int = 5
+    ) -> list[Memory]:
+        """检索记忆"""
+        if memory_types is None:
+            memory_types = [MemoryType.PERSISTENT]
+
+        type_values = [t.value for t in memory_types]
+        placeholders = ",".join("?" * len(type_values))
+
+        cursor = self.conn.execute(f"""
+            SELECT id, memory_type, content, embedding, created_at
+            FROM memories
+            WHERE workspace_id = ? AND memory_type IN ({placeholders})
+            ORDER BY created_at DESC
+            LIMIT ?
+        """, [workspace_id] + type_values + [top_k])
+
+        return [
+            Memory(
+                id=row[0],
+                memory_type=MemoryType(row[1]),
+                content=row[2],
+                embedding=pickle.loads(row[3]) if row[3] else [],
+                created_at=row[4],
+                workspace_id=workspace_id
+            )
+            for row in cursor
+        ]
+
+    def build_context(self, query: str, workspace_id: str) -> str:
+        """构建完整上下文"""
+        query_embedding = self._simple_embed(query)
+
+        # 检索持久记忆 + 会话记忆
+        persistent = self.retrieve(query_embedding, workspace_id, [MemoryType.PERSISTENT], top_k=3)
+        session = self.retrieve(query_embedding, workspace_id, [MemoryType.SESSION], top_k=5)
+
+        parts = ["## 相关记忆"]
+        for m in persistent + session:
+            parts.append(f"- [{m.memory_type.value}] {m.content[:200]}")
+
+        return "\n".join(parts)
+
+    def _simple_embed(self, text: str) -> list[float]:
+        """简化版 embedding"""
+        import hashlib
+        h = hashlib.sha256(text.encode()).digest()
+        return [b / 255.0 for b in h[:32]]
+```
+
 ## LoopForge 的记忆设计
 
 作为本地优先的 Agent OS，LoopForge 实现了完整的多层记忆系统：
@@ -238,6 +357,6 @@ Agent 的记忆不是"记住所有事情"，而是**在对的时刻想起对的
 
 **相关链接**
 
-- [LoopForge 记忆模块源码](../explanation/memory-subsystem.md)
-- [向量检索最佳实践](../how-to/vector-search.md)
+- [LoopForge 记忆模块源码（rexos-memory）](https://github.com/rexleimo/LoopForge/tree/main/meos/crates/rexos-memory)
+- [LoopForge 概念与记忆模型](../explanation/concepts.md)
 - [Harness 教程：长任务执行](../tutorials/harness-long-task.md)

docs-site/blog/editorial-calendar.md

@@ -0,0 +1,42 @@
+# Editorial Calendar
+
+Use this template to ship one reproducible, execution-focused post every week.
+
+## Weekly Cadence
+
+1. Monday: pick one real user pain point.
+2. Tuesday: collect reproducible commands and evidence screenshots.
+3. Wednesday: draft the post (problem -> commands -> verification -> pitfalls).
+4. Thursday: run internal technical review.
+5. Friday: publish and sync updates to changelog/community channels.
+
+## Post Template
+
+~~~markdown
+# [Outcome-first title]
+
+## Problem
+- Who is blocked?
+- Where does it fail in a real workflow?
+
+## Reproduce
+$ [minimal reproducible commands]
+
+## Fix Pattern
+$ [stable command sequence]
+
+## Verification
+- Which command proves success?
+- Which artifact proves success?
+
+## Common Pitfalls
+- Typical configuration mistakes
+- Site/provider instability notes
+~~~
+
+## Suggested 4-Week Topics
+
+1. Layered smoke strategy for stable vs dynamic websites.
+2. One-command onboarding and failure recovery.
+3. Provider routing strategy for cost vs performance.
+4. Harness checkpoint-driven long-task refactoring.

docs-site/blog/mcp-integration-guide.md

@@ -159,6 +159,107 @@ loopforge agent run \
   --prompt "阅读 src/main.rs，分析代码结构，写一份500字的架构说明"
 ```
 
+## 进阶：自定义 MCP Server
+
+如果官方 Server 不满足需求，可以自己写：
+
+### 1. Python 实现
+
+```python
+# my_mcp_server.py
+from mcp.server import Server
+from mcp.server.stdio import stdio_server
+from pydantic import AnyUrl
+import json
+
+app = Server("my-custom-server")
+
+@app.list_resources()
+async def list_resources():
+    return [
+        Resource(
+            uri=AnyUrl("myapp://config"),
+            name="app_config",
+            description="Application configuration"
+        )
+    ]
+
+@app.read_resource()
+async def read_resource(uri: AnyUrl):
+    if uri == "myapp://config":
+        with open("config.json") as f:
+            return json.load(f)
+    raise ValueError(f"Unknown resource: {uri}")
+
+@app.list_tools()
+async def list_tools():
+    return [
+        Tool(
+            name="analyze_code",
+            description="Analyze Python code for issues",
+            inputSchema={
+                "type": "object",
+                "properties": {
+                    "file_path": {"type": "string"}
+                },
+                "required": ["file_path"]
+            }
+        )
+    ]
+
+@app.call_tool()
+async def call_tool(name: str, arguments: dict):
+    if name == "analyze_code":
+        # 实现代码分析逻辑
+        return analyze_python(arguments["file_path"])
+    raise ValueError(f"Unknown tool: {name}")
+
+if __name__ == "__main__":
+    stdio_server.run(app)
+```
+
+### 2. 注册到 LoopForge
+
+```json
+{
+  "servers": {
+    "my-custom": {
+      "command": "python",
+      "args": ["/path/to/my_mcp_server.py"]
+    }
+  }
+}
+```
+
+### 3. 更多官方 MCP Server
+
+| Server | 功能 |
+|--------|------|
+| `@modelcontextprotocol/server-filesystem` | 文件系统操作 |
+| `@modelcontextprotocol/server-git` | Git 操作 |
+| `@modelcontextprotocol/server-github` | GitHub API |
+| `@modelcontextprotocol/server-brave-search` | 网页搜索 |
+| `@modelcontextprotocol/server-sqlite` | SQLite 数据库 |
+| `@modelcontextprotocol/server-postgres` | PostgreSQL |
+| `@modelcontextprotocol/server-memory` | 知识图谱记忆 |
+
+## 常见问题
+
+### Q: MCP 和 API 有什么区别？
+
+API 是"你问我答"，MCP 是"我知我能"。MCP 让 AI 知道自己有什么工具可用，不需要每次都通过 prompt 告知。
+
+### Q: 安全吗？
+
+MCP 本身是协议，安全取决于 Server 实现。建议：
+- 本地开发用 filesystem server
+- 生产环境用细粒度权限控制
+- 不推荐直接暴露 MCP Server 到公网
+
+### Q: 支持流式输出吗？
+
+支持。MCP 基于 JSON-RPC 2.0，支持 `server->client` 通知，可以实现实时进度更新。
+
 ## 总结
 
 MCP 不仅仅是一个协议——它是 AI 从"问答助手"进化到"执行引擎"的关键一步。