auto-coder 1.0.0__py3-none-any.whl → 2.0.1__py3-none-any.whl

autocoder/common/rulefiles/core/selector.py

@@ -0,0 +1,805 @@
+# -*- coding: utf-8 -*-
+"""
+RuleSelector
+
+基于 LLM 的智能规则选择器，支持规则摘要生成和索引创建。
+"""
+
+import os
+import subprocess
+import tempfile
+import concurrent.futures
+from typing import Dict, List, Optional, Union
+from loguru import logger
+
+# 导入第三方依赖
+import byzerllm
+from autocoder.common import AutoCoderArgs
+
+# 导入数据模型
+from ..models.rule_file import RuleFile
+from ..models.rule_relevance import RuleRelevance
+from ..models.summary import AlwaysApplyRuleSummary
+from ..models.index import ConditionalRulesIndex
+from ..models.init_rule import InitRule
+
+
+
+
+class RuleSelector:
+    """
+    根据LLM的判断和规则元数据选择适用的规则。
+    """
+    def __init__(self, llm: Optional[Union[byzerllm.ByzerLLM, byzerllm.SimpleByzerLLM]], args: Optional[AutoCoderArgs] = None):
+        """
+        初始化RuleSelector。
+
+        Args:
+            llm: ByzerLLM 实例，用于判断规则是否适用。如果为 None，则只选择 always_apply=True 的规则。
+            args: 传递给 Agent 的参数，可能包含用于规则选择的上下文信息。            
+        """
+        self.llm = llm
+        self.args = args
+        
+        
+
+    @byzerllm.prompt()
+    def _build_selection_prompt(self, rule: RuleFile, context: str = "") -> str:
+        """
+        判断规则是否适用于当前任务。
+
+        规则描述:
+        {{ rule.description }}
+
+        规则内容摘要 (前200字符):
+        {{ rule.content[:200] }}
+
+        {% if context %}
+        任务上下文:
+        {{ context }}
+        {% endif %}
+
+        基于以上信息，判断这条规则 (路径: {{ rule.file_path }}) 是否与当前任务相关并应该被应用？
+        
+        请以JSON格式返回结果:
+        ```json
+        {
+            "is_relevant": true或false,
+            "reason": "判断理由"
+        }
+        ```
+        """
+        # 注意：确保 rule 对象和 context 字典能够被 Jinja2 正确访问。
+        # Pydantic模型可以直接在Jinja2中使用其属性。
+        return {
+            "rule": rule,
+            "context": context
+        } # type: ignore
+        
+    def _evaluate_rule(self, rule: RuleFile, context: str) -> tuple[RuleFile, bool, Optional[str]]:
+        """
+        评估单个规则是否适用于当前上下文。
+        
+        Args:
+            rule: 要评估的规则
+            context: 上下文信息
+            
+        Returns:
+            tuple: (规则, 是否选中, 理由)
+        """
+        # 如果规则设置为总是应用，直接返回选中
+        if rule.always_apply:
+            return (rule, True, "规则设置为总是应用")
+            
+        # 如果没有LLM，无法评估non-always规则
+        if self.llm is None:
+            return (rule, False, "未提供LLM，无法评估non-always规则")
+            
+        try:
+            prompt = self._build_selection_prompt.prompt(rule=rule, context=context)
+            logger.debug(f"为规则 '{os.path.basename(rule.file_path)}' 生成的判断 Prompt (片段): {prompt[:200]}...")
+            
+            result = None
+            try:
+                # 使用with_return_type方法获取结构化结果
+                result = self._build_selection_prompt.with_llm(self.llm).with_return_type(RuleRelevance).run(rule=rule, context=context)
+                if result and result.is_relevant:
+                    return (rule, True, result.reason)
+                else:
+                    return (rule, False, result.reason if result else "未提供理由")
+            except Exception as e:
+                logger.warning(f"LLM 未能为规则 '{os.path.basename(rule.file_path)}' 提供有效响应: {e}")
+                return (rule, False, f"LLM评估出错: {str(e)}")
+                
+        except Exception as e:
+            logger.error(f"评估规则 '{os.path.basename(rule.file_path)}' 时出错: {e}", exc_info=True)
+            return (rule, False, f"评估过程出错: {str(e)}")
+
+    def select_rules(self, context: str) -> List[RuleFile]:
+        """
+        选择适用于当前上下文的规则。使用线程池并发评估规则。
+
+        Args:
+            context: 包含用于规则选择的上下文信息 (例如，用户指令、目标文件等)。
+
+        Returns:
+            List[RuleFile]: 选定的规则列表。
+        """
+        # 导入函数以避免循环依赖
+        from ..api import get_parsed_rules
+        
+        rules = get_parsed_rules()
+        selected_rules: List[RuleFile] = []
+        logger.info(f"开始选择规则，总规则数: {len(rules)}")
+        
+        # 预先分类处理always_apply规则
+        always_apply_rules = []
+        need_llm_rules = []
+        
+        for rule in rules:
+            if rule.always_apply:
+                always_apply_rules.append(rule)
+            elif self.llm is not None:
+                need_llm_rules.append(rule)
+        
+        # 添加always_apply规则
+        for rule in always_apply_rules:
+            selected_rules.append(rule)
+            logger.debug(f"规则 '{os.path.basename(rule.file_path)}' (AlwaysApply=True) 已自动选择。")
+            
+        # 如果没有需要LLM评估的规则，直接返回结果
+        if not need_llm_rules:
+            logger.info(f"规则选择完成，选中规则数: {len(selected_rules)}")
+            return selected_rules
+            
+        # 使用线程池并发评估规则
+        with concurrent.futures.ThreadPoolExecutor() as executor:
+            # 提交所有评估任务
+            future_to_rule = {
+                executor.submit(self._evaluate_rule, rule, context): rule 
+                for rule in need_llm_rules
+            }
+            
+            # 收集评估结果
+            for future in concurrent.futures.as_completed(future_to_rule):
+                rule, is_selected, reason = future.result()
+                if is_selected:
+                    selected_rules.append(rule)
+                    logger.info(f"规则 '{os.path.basename(rule.file_path)}' (AlwaysApply=False) 已被 LLM 选择，原因: {reason}")
+                else:
+                    logger.debug(f"规则 '{os.path.basename(rule.file_path)}' (AlwaysApply=False) 未被 LLM 选择，原因: {reason}")
+                    
+        logger.info(f"规则选择完成，选中规则数: {len(selected_rules)}")
+        return selected_rules
+
+    def get_selected_rules_content(self, context: str) -> Dict[str, str]:
+        """
+        获取选定规则的文件路径和内容字典。
+
+        Args:
+            context: 传递给 select_rules 的上下文。
+
+        Returns:
+            Dict[str, str]: 选定规则的 {file_path: content} 字典。
+        """
+        selected_rules = self.select_rules(context=context)
+        # 使用 os.path.basename 获取文件名作为 key，如果需要的话
+        # return {os.path.basename(rule.file_path): rule.content for rule in selected_rules}
+        # 保持 file_path 作为 key
+        return {rule.file_path: rule.content for rule in selected_rules}
+
+    @byzerllm.prompt()
+    def _build_always_apply_summary_prompt(self, rules: List[RuleFile]) -> str:
+        """
+        Merge all always-apply rules into a unified rule description.
+
+        Below is the list of all always-apply rules ({{ rules|length }} rules in total):
+
+        {% for rule in rules %}
+        ## Rule {{ loop.index }}: {{ rule.file_path }}
+        {% if rule.description %}
+        **Description**: {{ rule.description }}
+        {% endif %}
+        {% if rule.globs %}
+        **Applicable Files**: {{ rule.globs|join(', ') }}
+        {% endif %}
+
+        **Content**:
+        {{ rule.content }}
+
+        ---
+        {% endfor %}
+
+        Please merge all the above rules into a unified and concise rule description. Requirements:
+        1. Keep all important rule points
+        2. Remove duplicate content while preserving complementary information
+        3. Organize content by logical grouping
+        4. Generate a unified rule document that is easy to understand and follow
+
+        ***Do not include any extraneous content, please start outputting the rule content directly.***
+        ```
+        """
+        return {"rules": rules}  # type: ignore
+
+    @byzerllm.prompt()
+    def _build_conditional_rules_index_prompt(self, rules: List[RuleFile]) -> str:
+        """
+        Generate an index directory for all conditional rules.
+
+        Below is the list of all conditional rules ({{ rules|length }} rules in total):
+
+        {% for rule in rules %}
+        ## Rule {{ loop.index }}: {{ rule.file_path }}
+        {% if rule.description %}
+        **Description**: {{ rule.description }}
+        {% endif %}
+        {% if rule.globs %}
+        **Applicable Files**: {{ rule.globs|join(', ') }}
+        {% endif %}
+
+        **Content**:
+        {{ rule.content }}
+
+        ---
+        {% endfor %}
+
+        Please generate a detailed index directory for all the above conditional rules. Requirements:        
+        1. Provide concise descriptions and usage explanations for each rule to help users understand the purpose of the rules
+        2. Explain when these rules should be selected and used
+        3. Generate an easy-to-navigate and searchable directory structure that must include file paths        
+
+        ***Do not include any extraneous content, please start outputting the rule content directly.***
+        """
+        return {"rules": rules}  # type: ignore
+
+    def generate_always_apply_summary(self) -> Optional[AlwaysApplyRuleSummary]:
+        """
+        生成所有必须应用规则的合并摘要。
+
+        Returns:
+            AlwaysApplyRuleSummary: 合并后的规则摘要，如果没有 LLM 则返回 None
+        """
+        if self.llm is None:
+            logger.warning("未提供 LLM，无法生成必须应用规则摘要")
+            return None
+
+        # 导入函数以避免循环依赖
+        from ..api import get_parsed_rules
+        
+        # 获取所有解析后的规则
+        all_rules = get_parsed_rules()
+        always_apply_rules = [rule for rule in all_rules if rule.always_apply]
+        
+        if not always_apply_rules:
+            logger.info("未找到必须应用的规则")
+            return AlwaysApplyRuleSummary(
+                summary="未找到必须应用的规则。",
+                rule_count=0,
+                covered_areas=[]
+            )
+
+
+
+        logger.info(f"开始生成必须应用规则摘要，规则数量: {len(always_apply_rules)}")
+
+        try:
+            # 由于 prompt 现在不返回 JSON，我们需要直接获取字符串结果
+            result_str = self._build_always_apply_summary_prompt.with_llm(self.llm).run(rules=always_apply_rules)
+            if result_str:
+                # 创建摘要对象
+                result = AlwaysApplyRuleSummary(
+                    summary=result_str,
+                    rule_count=len(always_apply_rules),
+                    covered_areas=[]  # 暂时为空，可以后续通过 LLM 分析生成
+                )
+                
+
+                
+                logger.info(f"成功生成必须应用规则摘要")
+                return result
+            else:
+                logger.warning("LLM 未能生成有效的规则摘要")
+                return None
+        except Exception as e:
+            logger.error(f"生成必须应用规则摘要时出错: {e}", exc_info=True)
+            return None
+
+    def generate_conditional_rules_index(self) -> Optional[ConditionalRulesIndex]:
+        """
+        生成所有条件规则的索引目录。
+
+        Returns:
+            ConditionalRulesIndex: 条件规则索引，如果没有 LLM 则返回 None
+        """
+        if self.llm is None:
+            logger.warning("未提供 LLM，无法生成条件规则索引")
+            return None
+
+        # 导入函数以避免循环依赖
+        from ..api import get_parsed_rules
+        
+        # 获取所有解析后的规则
+        all_rules = get_parsed_rules()
+        conditional_rules = [rule for rule in all_rules if not rule.always_apply]
+        
+        if not conditional_rules:
+            logger.info("未找到条件规则")
+            return ConditionalRulesIndex(
+                index_content="未找到条件规则。",
+                rule_count=0,
+                categories=[]
+            )
+
+
+
+        logger.info(f"开始生成条件规则索引，规则数量: {len(conditional_rules)}")
+
+        try:
+            # 由于 prompt 现在不返回 JSON，我们需要直接获取字符串结果
+            result_str = self._build_conditional_rules_index_prompt.with_llm(self.llm).run(rules=conditional_rules)
+            if result_str:
+                # 创建索引对象
+                result = ConditionalRulesIndex(
+                    index_content=result_str,
+                    rule_count=len(conditional_rules),
+                    categories=[]  # 暂时为空，可以后续通过 LLM 分析生成
+                )
+                
+
+                
+                logger.info(f"成功生成条件规则索引")
+                return result
+            else:
+                logger.warning("LLM 未能生成有效的规则索引")
+                return None
+        except Exception as e:
+            logger.error(f"生成条件规则索引时出错: {e}", exc_info=True)
+            return None
+    
+
+
+    def generate_init_rule(self, project_root: Optional[str] = None) -> Optional[InitRule]:
+        """
+        使用 subagent 机制探索项目结构并生成初始化规则。
+
+        Args:
+            project_root: 项目根目录路径，如果为 None 则使用当前目录
+
+        Returns:
+            InitRule: 生成的初始化规则，如果失败则返回 None
+        """
+        if project_root is None:
+            project_root = os.getcwd()
+
+        logger.info(f"使用 subagent 机制开始生成初始化规则: {project_root}")
+
+        try:
+            # 探索项目结构
+            project_info = self._explore_project_structure(project_root)
+            logger.info(f"项目探索完成，检测到技术栈: {project_info['technologies']}")
+
+            # 获取模型名称
+            model_name = self._get_model_name()
+            if not model_name:
+                logger.error("无法获取模型名称，无法使用 subagent 机制")
+                return None
+
+            # 使用 subagent 机制生成规则
+            success = self._generate_init_rule_with_subagent(project_info, model_name, project_root)
+            
+            if not success:
+                logger.error("subagent 执行失败")
+                return None
+
+            # 检查生成的文件并读取内容
+            rules_dir = self._find_rules_directory(project_root)
+            if not rules_dir:
+                logger.error("未找到 .autocoderrules 目录")
+                return None
+
+            init_file_path = os.path.join(rules_dir, "init.md")
+            if not os.path.exists(init_file_path):
+                logger.error(f"未找到生成的 init.md 文件: {init_file_path}")
+                return None
+
+            # 读取生成的内容
+            with open(init_file_path, 'r', encoding='utf-8') as f:
+                content = f.read()
+
+            if not content.strip():
+                logger.error("生成的 init.md 文件为空")
+                return None
+
+            # 确定项目类型
+            project_type = self._determine_project_type(project_info['technologies'])
+            
+            # 提取命令列表
+            commands = self._extract_commands_from_content(content)
+            
+            # 创建 InitRule 对象
+            result = InitRule(
+                content=content,
+                project_type=project_type,
+                commands=commands,
+                technologies=project_info['technologies'],
+                file_path=init_file_path
+            )
+            
+            logger.info(f"成功使用 subagent 生成初始化规则，项目类型: {project_type}")
+            return result
+
+        except Exception as e:
+            logger.error(f"使用 subagent 生成初始化规则时出错: {e}", exc_info=True)
+            return None
+
+    def _explore_project_structure(self, project_root: str) -> Dict:
+        """探索项目结构，检测技术栈和配置文件"""
+        project_info = {
+            'root_files': [],
+            'main_directories': [],
+            'technologies': [],
+            'package_json': None,
+            'setup_py': None,
+            'requirements_txt': None,
+            'go_mod': None,
+            'pyproject_toml': None,
+            'tsconfig_json': None,
+        }
+
+        try:
+            # 获取根目录文件
+            if os.path.exists(project_root):
+                root_items = os.listdir(project_root)
+                
+                # 分离文件和目录
+                for item in root_items:
+                    item_path = os.path.join(project_root, item)
+                    if os.path.isfile(item_path):
+                        project_info['root_files'].append(item)
+                    elif os.path.isdir(item_path) and not item.startswith('.'):
+                        project_info['main_directories'].append(item)
+
+                # 检测配置文件并读取内容
+                config_files = {
+                    'package.json': 'package_json',
+                    'setup.py': 'setup_py', 
+                    'requirements.txt': 'requirements_txt',
+                    'go.mod': 'go_mod',
+                    'pyproject.toml': 'pyproject_toml',
+                    'tsconfig.json': 'tsconfig_json',
+                }
+
+                for filename, key in config_files.items():
+                    file_path = os.path.join(project_root, filename)
+                    if os.path.exists(file_path):
+                        try:
+                            with open(file_path, 'r', encoding='utf-8') as f:
+                                content = f.read()
+                                project_info[key] = content[:1000]  # 限制内容长度
+                        except Exception as e:
+                            logger.warning(f"读取 {filename} 时出错: {e}")
+
+                # 检测技术栈
+                project_info['technologies'] = self._detect_technologies(project_info)
+
+        except Exception as e:
+            logger.error(f"探索项目结构时出错: {e}")
+
+        return project_info
+
+    def _detect_technologies(self, project_info: Dict) -> List[str]:
+        """根据配置文件检测技术栈"""
+        technologies = []
+
+        # Python
+        if project_info['setup_py'] or project_info['requirements_txt'] or project_info['pyproject_toml']:
+            technologies.append('Python')
+
+        # JavaScript/TypeScript
+        if project_info['package_json']:
+            technologies.append('JavaScript')
+            try:
+                import json
+                package_data = json.loads(project_info['package_json'])
+                dependencies = {
+                    **package_data.get('dependencies', {}),
+                    **package_data.get('devDependencies', {})
+                }
+                
+                if 'typescript' in dependencies or project_info['tsconfig_json']:
+                    technologies.append('TypeScript')
+                
+                if 'react' in dependencies:
+                    technologies.append('React')
+                
+                if 'vue' in dependencies:
+                    technologies.append('Vue')
+                
+                if 'express' in dependencies:
+                    technologies.append('Node.js')
+                    
+            except Exception as e:
+                logger.warning(f"解析 package.json 时出错: {e}")
+
+        # Go
+        if project_info['go_mod']:
+            technologies.append('Go')
+
+        # 其他检测
+        if 'src' in project_info['main_directories']:
+            if not technologies:  # 如果还没检测到技术栈，根据目录结构猜测
+                technologies.append('General')
+
+        return technologies if technologies else ['Unknown']
+
+    def _determine_project_type(self, technologies: List[str]) -> str:
+        """根据技术栈确定项目类型"""
+        if 'React' in technologies:
+            return 'React Application'
+        elif 'Vue' in technologies:
+            return 'Vue Application'
+        elif 'TypeScript' in technologies:
+            return 'TypeScript Project'
+        elif 'JavaScript' in technologies:
+            return 'JavaScript Project'
+        elif 'Python' in technologies:
+            return 'Python Project'
+        elif 'Go' in technologies:
+            return 'Go Project'
+        else:
+            return 'General Project'
+
+    def _extract_commands_from_content(self, content: str) -> List[str]:
+        """从生成的内容中提取命令列表"""
+        commands = []
+        lines = content.split('\n')
+        in_commands_section = False
+        
+        for line in lines:
+            line = line.strip()
+            if line.lower().startswith('# bash commands'):
+                in_commands_section = True
+                continue
+            elif line.startswith('#') and in_commands_section:
+                in_commands_section = False
+            elif in_commands_section and line.startswith('-'):
+                # 提取命令名称
+                command_part = line[1:].strip()
+                if ':' in command_part:
+                    command_name = command_part.split(':')[0].strip()
+                    commands.append(command_name)
+        
+        return commands
+
+    def _get_model_name(self) -> Optional[str]:
+        """
+        获取当前可用的模型名称
+        
+        Returns:
+            当前模型名称，如果没有则返回 None
+        """
+        try:
+            # 尝试从 args 中获取模型信息
+            if self.args and hasattr(self.args, 'model') and self.args.model:
+                return self.args.model
+            
+            # 尝试从 llm 实例获取模型名称
+            if self.llm:
+                if hasattr(self.llm, 'default_model_name') and self.llm.default_model_name:
+                    return self.llm.default_model_name
+                elif hasattr(self.llm, 'model_name') and self.llm.model_name:
+                    return self.llm.model_name
+                elif hasattr(self.llm, 'model') and self.llm.model:
+                    return self.llm.model
+            
+            # 尝试从环境变量获取
+            env_model = os.getenv('AUTO_CODER_MODEL')
+            if env_model:
+                return env_model
+            
+            # 默认模型
+            return "v3_chat"
+            
+        except Exception as e:
+            logger.warning(f"获取模型名称时出错: {str(e)}")
+            return "v3_chat"
+
+    def _generate_init_rule_with_subagent(self, project_info: Dict, model_name: str, project_root: str) -> bool:
+        """
+        使用 subagent 机制生成初始化规则
+        
+        Args:
+            project_info: 项目信息
+            model_name: 模型名称
+            project_root: 项目根目录
+            
+        Returns:
+            是否成功
+        """
+        try:
+            # 创建临时文件
+            with tempfile.NamedTemporaryFile(mode='w', suffix='.md', delete=False, encoding='utf-8') as temp_file:
+                temp_file_path = temp_file.name
+                
+                # 构建指令内容
+                instruction_content = self._build_subagent_instruction(project_info, project_root)
+                temp_file.write(instruction_content)
+                
+            logger.info(f"创建临时指令文件: {temp_file_path}")
+            
+            try:
+                # 切换到项目根目录执行命令
+                original_cwd = os.getcwd()
+                os.chdir(project_root)
+                
+                # 构建并执行 auto-coder.run 命令
+                command = f"cat {temp_file_path} | auto-coder.run --model {model_name}"
+                logger.info(f"执行 subagent 命令: {command}")
+                
+                # 执行命令
+                result = subprocess.run(
+                    command,
+                    shell=True,
+                    capture_output=True,
+                    text=True,
+                    timeout=300  # 5分钟超时
+                )
+                
+                if result.returncode == 0:
+                    logger.info("subagent 命令执行成功")
+                    if result.stdout:
+                        logger.info(f"命令输出: {result.stdout}")
+                    return True
+                else:
+                    logger.error(f"subagent 命令执行失败，返回码: {result.returncode}")
+                    if result.stderr:
+                        logger.error(f"错误输出: {result.stderr}")
+                    if result.stdout:
+                        logger.error(f"标准输出: {result.stdout}")
+                    return False
+                    
+            finally:
+                # 恢复原始工作目录
+                os.chdir(original_cwd)
+                
+                # 清理临时文件
+                try:
+                    os.unlink(temp_file_path)
+                    logger.info(f"已清理临时文件: {temp_file_path}")
+                except Exception as e:
+                    logger.warning(f"清理临时文件时出错: {e}")
+                    
+        except subprocess.TimeoutExpired:
+            logger.error("subagent 命令执行超时")
+            return False
+        except Exception as e:
+            logger.error(f"执行 subagent 命令时出错: {e}", exc_info=True)
+            return False
+
+    def _build_subagent_instruction(self, project_info: Dict, project_root: str) -> str:
+        """
+        构建 subagent 的指令内容
+        
+        Args:
+            project_info: 项目信息
+            project_root: 项目根目录
+            
+        Returns:
+            指令内容
+        """
+        # 构建项目信息描述
+        project_description = ""
+        
+        if project_info.get('package_json'):
+            project_description += f"\n**发现 package.json 文件**:\n```json\n{project_info['package_json'][:500]}...\n```\n"
+        
+        if project_info.get('setup_py'):
+            project_description += f"\n**发现 setup.py 文件**:\n```python\n{project_info['setup_py'][:500]}...\n```\n"
+            
+        if project_info.get('requirements_txt'):
+            project_description += f"\n**发现 requirements.txt 文件**:\n```\n{project_info['requirements_txt'][:300]}...\n```\n"
+            
+        if project_info.get('go_mod'):
+            project_description += f"\n**发现 go.mod 文件**:\n```\n{project_info['go_mod'][:300]}...\n```\n"
+            
+        if project_info.get('pyproject_toml'):
+            project_description += f"\n**发现 pyproject.toml 文件**:\n```toml\n{project_info['pyproject_toml'][:500]}...\n```\n"
+            
+        if project_info.get('tsconfig_json'):
+            project_description += f"\n**发现 tsconfig.json 文件**:\n```json\n{project_info['tsconfig_json'][:300]}...\n```\n"
+        
+        root_files = ', '.join(project_info.get('root_files', [])[:10])
+        main_directories = ', '.join(project_info.get('main_directories', [])[:10])
+        technologies = ', '.join(project_info.get('technologies', []))
+        
+        instruction = f"""请分析当前项目结构并在 .autocoderrules 目录下生成一个 init.md 文件，包含项目的开发规则和指导。
+
+## 项目信息
+
+**项目根目录**: {project_root}
+**根目录文件**: {root_files}
+**主要目录**: {main_directories}
+**检测到的技术栈**: {technologies}
+
+{project_description}
+
+## 任务要求
+
+请根据上述项目信息，在 `.autocoderrules` 目录下创建一个 `init.md` 文件，内容应该包括：
+
+### 1. Bash commands
+列出常用的构建、测试、运行命令，格式如下：
+```
+# Bash commands
+- command_name: 命令描述
+- another_command: 另一个命令描述
+```
+
+### 2. Code style  
+根据检测到的技术栈，提供相应的编码规范，例如：
+```
+# Code style
+- 使用一致的代码格式化规则
+- 遵循对应语言的最佳实践
+```
+
+### 3. Workflow
+提供开发工作流建议：
+```
+# Workflow
+- 开发流程建议
+- 测试和部署流程
+```
+
+### 4. Technology-specific rules
+针对具体技术栈的特殊规则：
+```
+# Technology-specific rules
+- 技术栈特定的规则
+- 框架相关的最佳实践
+```
+
+## 注意事项
+
+1. 如果 `.autocoderrules` 目录不存在，请先创建该目录
+2. 根据实际检测到的技术栈生成相应的规则
+3. 确保生成的规则实用且具体
+4. 命令应该是项目中实际可能用到的
+5. 规则应该帮助开发者快速上手项目开发
+
+请立即开始分析项目并生成 init.md 文件。
+"""
+        
+        return instruction
+
+    def _find_rules_directory(self, project_root: str) -> Optional[str]:
+        """
+        查找现有的 .autocoderrules 目录
+        
+        Args:
+            project_root: 项目根目录
+            
+        Returns:
+            找到的规则目录路径，如果没找到则返回 None
+        """
+        try:
+            # 按优先级顺序查找
+            possible_dirs = [
+                os.path.join(project_root, ".autocoderrules"),
+                os.path.join(project_root, ".auto-coder", ".autocoderrules"),
+                os.path.join(project_root, ".auto-coder", "autocoderrules"),
+            ]
+            
+            for rules_dir in possible_dirs:
+                if os.path.exists(rules_dir) and os.path.isdir(rules_dir):
+                    logger.info(f"找到现有的规则目录: {rules_dir}")
+                    return rules_dir
+                    
+            # 如果没有找到现有目录，返回默认的第一个位置
+            default_dir = possible_dirs[0]
+            logger.info(f"未找到现有规则目录，返回默认位置: {default_dir}")
+            return default_dir
+            
+        except Exception as e:
+            logger.error(f"查找规则目录时出错: {e}")
+            return None