中文 English

This repository maintains a skill package that guides an agent to generate Adnc CRUD code from SQL DDL. It is for authoring, refining, and validating the skill itself rather than building the business application.

Purpose

• Maintain the generation contract in SKILL.md
• Keep templates and rule documents aligned with the target Adnc repository
• Use mirrored source snapshots under references\src as examples for generation behavior
• Generate real output under .\src, never under references\src

Repository layout

Path	Role
SKILL.md	Main skill contract: triggers, required inputs, workflow, output scope, and generation rules
INSTRUCTIONS.md	Repository purpose and maintenance guidance
assets\templates.md	Skeletons for generated entities, DTOs, validators, services, controllers, and shared updates
references\ddl-mapping.md	Mapping rules from SQL DDL to Adnc repository, application, and API code
references\project-conventions.md	Required reference files and repo-specific conventions
references\src	Mirrored source snapshots for pattern matching only

What the skill generates

For a normal business table, the skill is expected to generate or update:

• .\src\Repository\Entities\{Entity}.cs
• .\src\Repository\Entities\Config\{Entity}Config.cs
• .\src\Repository\EntityInfo.cs
• .\src\Repository\EntityConsts.cs when reusable length constants are needed
• .\src\Application\Contracts\Dtos\{Entity}\*
• .\src\Application\Contracts\Interfaces\I{Entity}Service.cs
• .\src\Application\Services\{Entity}Service.cs
• .\src\Application\MapperProfile.cs
• .\src\Api\Controllers\{Entity}Controller.cs
• .\src\Api\PermissionConsts.cs

Pure relation tables such as *_relation should skip full CRUD by default unless the user explicitly asks for it.

Required inputs

The skill should only run when the user provides:

1. A DDL file or CREATE TABLE statements for the target tables
2. The namespace prefix, for example Adnc.Demo.Admin

For the fastest and most reliable path, the DDL should also include:

• column COMMENT values
• explicit UNIQUE keys or unique indexes
• explicit isdeleted and rowversion columns when those behaviors are required

If the DDL is incomplete, the skill must stop and ask for the real table definition instead of guessing from partial SQL or seed data.

Direct usage in your team

There is no separate installer for this skill package right now. The most practical internal setup is to use this repository directly with GitHub Copilot CLI:

1. Install GitHub Copilot CLI.
2. Clone this repository.
3. Open this repository in your terminal.
4. Launch copilot.

Example setup:

winget install GitHub.Copilot
git clone https://github.com/AlphaYu/adnc-skill-ddl-codegen.git
cd adnc-skill-ddl-codegen
copilot

This means the repo is currently used as a shared skill repository rather than a one-click marketplace package.

How to use

After launching copilot in this repository, use the skill in a single prompt.

Recommended flow:

1. Paste one or more target CREATE TABLE statements.
2. State the namespace prefix explicitly.
3. Mention relation-table CRUD only when you want to override the default skip behavior.
4. Add scope limits only when you want less than the default full CRUD output.

How to write prompts

Good prompts should be short but complete. A useful prompt usually contains:

• the action: generate, update, or regenerate
• the target tables
• the DDL itself or a clear DDL source selected by the user
• the namespace prefix
• any non-default requirement, such as relation-table CRUD or limited output scope

Recommended prompt pattern:

Generate Adnc CRUD code for these tables.
Namespace prefix: <Your.Namespace.Prefix>
Special requirements: <only if needed>

<CREATE TABLE ...>

Prompt writing tips:

• Include the real CREATE TABLE statement instead of describing columns in prose.
• Include column comments when you want accurate entity property XML docs.
• Include unique keys or indexes when uniqueness checks should be generated in services.
• Explicitly say generate relation-table CRUD when the table is a pure join table and you do want full CRUD.
• Do not ask the skill to infer schema from insert scripts, screenshots, or partial SQL.

Examples:

• Basic prompt:

  Generate Adnc CRUD code from the following DDL.
  Namespace prefix: Adnc.Demo.Admin
  
  CREATE TABLE sys_customer (
      id bigint NOT NULL,
      customer_code varchar(32) NOT NULL COMMENT 'Customer code',
      customer_name varchar(128) NOT NULL COMMENT 'Customer name',
      isdeleted bit NOT NULL DEFAULT 0 COMMENT 'Deletion flag',
      rowversion rowversion NOT NULL,
      PRIMARY KEY (id),
      UNIQUE KEY uk_customer_code (customer_code)
  );
  

• Full CRUD for a normal table: Generate Adnc Admin CRUD code for this table. Namespace prefix: Adnc.Demo.Admin
• Force CRUD for a relation table: Generate relation-table CRUD for this DDL. Namespace prefix: Adnc.Demo.Admin
• Restrict scope: Generate only repository and application layers for this DDL. Namespace prefix: Adnc.Demo.Admin

Core rules

1. Read references\.editorconfig, references\ddl-mapping.md, and references\project-conventions.md before changing generation behavior.
2. Keep SKILL.md, INSTRUCTIONS.md, assets\templates.md, and relevant references\*.md files aligned when rules change.
3. Generate code into .\src\Repository, .\src\Application, and .\src\Api; never write generated output into references\src.
4. Entity property <summary> text must use the DDL column comment verbatim, or fall back to the raw SQL column name when no comment is available.
5. Preserve IsDeleted with ISoftDelete when the DDL contains isdeleted, and preserve RowVersion with IConcurrency when the DDL contains rowversion.
6. Update shared files such as EntityInfo.cs, MapperProfile.cs, and PermissionConsts.cs in the same pass as per-entity generation.
7. After successful generation, do not stage files with git; the completion message must be exactly 生成成功.

Maintenance guidance

• Prefer changing rules and templates over editing generated examples
• Treat references\src as read-only reference material
• Keep defaults explicit so the skill can take the fast path without repeated confirmation
• Optimize for deterministic generation, clear naming, and complete shared-file updates

License

This project is licensed under the MIT License. See LICENSE for details.