🌐 代码知识图谱

源代码 → 词法分析 → 语法分析 → AST
         ↓
    提取实体：类、函数、变量
    提取关系：调用、继承、导入

节点（Nodes）：
  - FileNode: "src/main.py"
  - ClassNode: "UserController"
  - FunctionNode: "create_user"
  - VariableNode: "db_connection"

边（Edges）：
  - CONTAINS: File → Class
  - CALLS: FunctionA → FunctionB
  - IMPORTS: FileA → FileB
  - INHERITS: ClassB → ClassA

# OpenClaw 配置示例
{
  "runtime": "subagent",
  "task": "审查PR #123的代码变更",
  "attachments": [
    {
      "name": "code-knowledge-graph",
      "content": "pre-indexed-graph.json",
      "encoding": "base64"
    }
  ]
}

# ~/.openclaw/skills/codegraph/SKILL.md
When user asks to understand a codebase or review code:
1. Load pre-built knowledge graph from .codegraph/
2. Query for relevant nodes: "function X calls which functions?"
3. Return structured context to Agent
4. Agent uses this context to generate accurate responses

# 传统方式：塞5000行代码给Agent
# 知识图谱方式：只给相关的200行 + 结构化关系描述

上下文 = {
  "target_function": "process_payment",
  "related_functions": ["validate_card", "charge_api", "log_transaction"],
  "call_graph": "process_payment → validate_card → charge_api",
  "relevant_files": ["payment.py", "api.py", "models.py"]
}

import ast
import networkx as nx

class CodeKnowledgeGraph:
    def __init__(self):
        self.graph = nx.DiGraph()
    
    def parse_file(self, filepath):
        """解析Python文件并构建图谱"""
        with open(filepath, 'r') as f:
            tree = ast.parse(f.read())
        
        # 添加文件节点
        self.graph.add_node(filepath, type='file')
        
        # 遍历AST
        for node in ast.walk(tree):
            if isinstance(node, ast.FunctionDef):
                func_name = f"{filepath}::{node.name}"
                self.graph.add_node(func_name, type='function')
                self.graph.add_edge(filepath, func_name, relation='contains')
                
                # 查找函数调用
                for child in ast.walk(node):
                    if isinstance(child, ast.Call):
                        if hasattr(child.func, 'id'):
                            called_func = f"::{child.func.id}"
                            self.graph.add_node(called_func, type='function_called')
                            self.graph.add_edge(func_name, called_func, relation='calls')
    
    def query_callers(self, function_name):
        """查询谁调用了某个函数"""
        callers = []
        for node in self.graph.nodes():
            if self.graph.has_edge(node, function_name):
                callers.append(node)
        return callers

# 使用示例
ckg = CodeKnowledgeGraph()
ckg.parse_file("src/payment.py")
print(ckg.query_callers("process_payment"))

# skills/codegraph-helper/SKILL.md
"""
OpenClaw Skill: Code Knowledge Graph Helper
用途：为OpenClaw Agent提供代码库结构化理解能力
"""

import json
from pathlib import Path

def load_knowledge_graph(repo_path):
    """加载预构建的知识图谱"""
    graph_path = Path(repo_path) / ".codegraph" / "graph.json"
    if graph_path.exists():
        with open(graph_path, 'r') as f:
            return json.load(f)
    return None

def query_graph(graph, query_type, query_value):
    """查询知识图谱"""
    if query_type == "callers":
        # 查找调用者
        return [n for n in graph['edges'] if n['to'] == query_value]
    elif query_type == "callees":
        # 查找被调用者
        return [n for n in graph['edges'] if n['from'] == query_value]
    elif query_type == "impact":
        # 查找影响范围（传递闭包）
        return compute_impact(graph, query_value)

# OpenClaw Agent调用示例
# Agent: "修改UserController.update_user方法会有什么影响？"
# → 调用此skill → 返回结构化影响分析 → Agent生成准确回答

# 将知识图谱信息注入Agent prompt
def enhance_prompt_with_graph(user_query, graph, file_path):
    """用知识图谱增强Agent的上下文"""
    
    # 查询相关上下文
    related_nodes = query_relevant_nodes(graph, user_query)
    call_graph = extract_call_graph(graph, file_path)
    
    # 构建增强prompt
    enhanced_context = f"""
代码库知识图谱上下文：
- 目标文件：{file_path}
- 相关函数：{related_nodes['functions']}
- 调用关系：{call_graph}
- 依赖文件：{related_nodes['dependencies']}

用户问题：{user_query}

请基于以上结构化上下文回答问题，注意函数间的调用关系。
"""
    return enhanced_context