Skip to main content

AI Workflow

The AI development process is driven by two main workflows: nl-to-dsl and execute-from-dsl.

NL-to-DSL Workflow​

This phase focuses on translating your requirements into a technical specification.

  1. Input: A document or message containing natural language requirements.
  2. AI Action:
    • Extracts Entities and Relationships (Mermaid).
    • Identifies business rules and constraints.
    • Defines service method signatures.
    • Proposes API endpoints.
  3. Output:
    • A DSL file: .ai/dsl/modules/{module-name}.md.
    • An implementation checklist: .ai/tasks/{module-name}-tasks.md.

Execute-from-DSL Workflow​

This phase focuses on the actual code generation.

  1. Input: The approved DSL file and task list.
  2. AI Action:
    • Follows the "Quests" from the playbook sequentially.
    • Generates C# code for each project in the module.
    • Adheres to conventions (snake_case database, specific namespaces, etc.).
    • Wires up the module using the registration patterns.
  3. Output: Production-ready code across all layers.

Verification​

After code generation, the agent:

  1. Runs dotnet build to ensure compilation.
  2. Optionally runs migrations and integration tests.
  3. Provides you with the necessary integration code for your Program.cs.

Tips for Success​

  • Be Specific in Requirements: Mention data types, mandatory fields, and visual relationships.
  • Review the DSL Carefully: The DSL is the "source of truth". If it's wrong, the code will be wrong.
  • Check the Task Checklist: Use the generated task list to track progress during execution.