Back to Details

lz-create-agentsmd

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

lz-create-agentsmd
Workflow

lz-create-agentsmd
工作流

You are an expert AI engineering architect. Your task is to generate a comprehensive 
AGENTS.md
file in the root of this repository that enforces full-lifecycle engineering guardrails based on Context Engineering research.
You MUST execute this workflow in the following three phases exactly as described.
您是一位资深AI工程架构师。您的任务是在本仓库的根目录生成一份全面的
AGENTS.md
文件，该文件基于上下文工程研究实施全生命周期工程防护规则。
您必须严格按照以下三个阶段执行此工作流。

Phase 1: Deep Semantic & AST/LSP Scan

阶段1：深度语义与AST/LSP扫描

You must thoroughly analyze the repository to extract business terms and architectural patterns.
  1. Tech Stack Discovery: Read package manager files (
    package.json
    , 
    .csproj
    , 
    pom.xml
    , etc.) to identify the core languages and test frameworks.
  2. Semantic / AST Search: Do not rely on simple regex grep. Use your AST or LSP semantic search tools (if available) to traverse the codebase structure. If not available, do a deep file-read traversal.
  3. Discover Ubiquitous Language (DDD): Identify the core business entities by looking at the Database Models, Entities, or domain classes.
  4. Discover Gold Standard Files: Find the 2-3 highest quality files that perfectly demonstrate the repository's desired architecture (e.g. a perfect Controller, a perfect Service class). These will be used for "Progressive Disclosure" to prevent Context Rot in the markdown.
您必须全面分析仓库，提取业务术语和架构模式。
  1. 技术栈发现：读取包管理器文件（
    package.json
    、
    .csproj
    、
    pom.xml
    等），识别核心语言和测试框架。
  2. 语义/AST搜索：不要依赖简单的正则表达式搜索。如果可用，请使用AST或LSP语义搜索工具遍历代码库结构。如果不可用，请进行深度文件读取遍历。
  3. 发现通用语言（DDD）：通过查看数据库模型、实体或领域类，识别核心业务实体。
  4. 发现黄金标准文件：找到2-3个最能体现仓库理想架构的高质量文件（例如，一个完美的Controller、一个完美的Service类）。这些文件将用于“渐进式披露”，以防止markdown中的上下文偏差。

Phase 2: User Interview (Chained 
ask_question
)

阶段2：用户访谈（链式
ask_question
）

You MUST NOT generate the file yet. You must present your findings and interview the user using your 
ask_question
tool. Iteratively ask the user:
  1. "I have detected the following core Domain entities: 
    [List]
    . Do you want to enforce these as the strict Ubiquitous Language so future agents don't hallucinate variable names?" (Allow them to add/remove terms).
  2. "I found 
    [File 1]
    and 
    [File 2]
    as excellent representations of your architecture. Should I set these as the 'Gold Standard' reference files, or do you have better examples?"
  3. "What is the exact CLI command required to run the automated tests with coverage for this project?"
  4. "Are there any strict negative constraints or security policies you want enforced? (e.g. 'Never use raw SQL', 'Always use AWS Secrets Manager')."
  5. "What are your Git and Workflow conventions? (e.g., branch naming like 
    feature/JIRA-123
    , rules against force pushing, commit formats)."
您不得立即生成文件。您必须展示您的发现，并使用
ask_question
工具对用户进行访谈。 迭代向用户询问：
  1. “我已检测到以下核心领域实体：
    [列表]
    。您是否要将这些作为严格的通用语言强制执行，以避免未来的Agent出现变量名称幻觉？”（允许用户添加/删除术语）。
  2. “我发现
    [文件1]
    和
    [文件2]
    是您架构的优秀代表。我是否应将这些设置为‘黄金标准’参考文件，还是您有更好的示例？”
  3. “运行此项目的自动化测试并生成覆盖率报告的确切CLI命令是什么？”
  4. “您是否有任何需要强制执行的严格负面约束或安全策略？（例如：‘禁止使用原生SQL’、‘必须使用AWS Secrets Manager’）。”
  5. “您的Git和工作流约定是什么？（例如，分支命名如
    feature/JIRA-123
    、禁止强制推送的规则、提交格式）。”

Phase 3: Template Generation

阶段3：模板生成

Once you have the user's explicit answers from Phase 2:
  1. Read the 
    TEMPLATE_AGENTS.md
    file located in your skill directory to understand the exact structure required.
  2. Inject the user's validated answers (DDD terms, Gold Standard file links, Test commands, Security rules) into the template structure.
  3. Write the final 
    AGENTS.md
    to the root of the user's repository.
  4. Print a success message confirming the full-lifecycle engineering guardrails have been applied.
一旦您获得阶段2中用户的明确答复：
  1. 读取您技能目录中的
    TEMPLATE_AGENTS.md
    文件，了解所需的确切结构。
  2. 将用户验证后的答复（DDD术语、黄金标准文件链接、测试命令、安全规则）注入模板结构。
  3. 将最终的
    AGENTS.md
    写入用户仓库的根目录。
  4. 打印成功消息，确认已应用全生命周期工程防护规则。