Requirements Generation

Requirement Generation Slice

This feature manages the generation of system requirements and sub-system requirements using AI. It operates completely decoupled from the database, performing pure business logic driven by Wolverine messages.

Architectural Components

1. RequirementGenerationHandler

The entry point for the requirement feature worker.

• It consumes a RequirementGenerationRequested event.
• It uses the provided InputClaimCheckKey to retrieve a RequirementGenerationPayload from the S3 Claim-Check Store.
• It maps and applies the project’s LLM presets (AiPresetConfigs) to the IAIService.
• It delegates the core processing to the RequirementGenerationCore.
• Finally, it saves the generation result back to the Claim-Check Store and returns a RequirementGenerationCompleted message to Wolverine.

2. RequirementGenerationCore

The pure logic core that drives the AI generation.

• Deduplication: Ensures that generated requirements are unique by comparing normalized body text (Jaccard similarity > 80%).
• Formatting: Assigns unique, sequential IDs to the generated requirements based on the document section title (e.g., REQ-SEC-001).
• Prompt Rendering: Uses PromptLibrary to combine the system prompts with context data.
• Handling Source Fallback: If no source files are available, it defaults to validating and improving a single existing requirement.
• State Transformation: Packages the generated items into RequirementBlockCreate and RequirementBlockUpdate structures so the API knows exactly how to apply the changes to the relational database.

Generation Sequence

The following diagram illustrates how the Requirements feature processes a request from the API.

sequenceDiagram
    participant API as API Apply/Prepare
    participant Broker as RabbitMQ
    participant S3 as Claim-Check Store
    participant Handler as RequirementGenerationHandler
    participant Core as RequirementGenerationCore
    participant AI as IAIService

    API->>S3: PutAsync(RequirementGenerationPayload)
    API->>Broker: Publish RequirementGenerationRequested
    Broker->>Handler: Consume Message
    Handler->>S3: GetAsync(PayloadKey)
    Handler->>Core: GenerateAsync(Payload, IAIService)
    
    alt Has Source Files
        Core->>AI: GenerateAsync(SystemPrompt, Context)
        AI-->>Core: JSON Requirements Array
        Core->>Core: DeduplicateRequirements()
        Core->>Core: Assign Sequential IDs
        Core-->>Handler: RequirementGenerationResult (Updates & Creates)
    else No Source Files
        Core->>AI: ValidateSingle(ExistingRequirement)
        AI-->>Core: Improved Requirement Text
        Core-->>Handler: RequirementGenerationResult (Single Update)
    end
    
    Handler->>S3: PutAsync(RequirementGenerationResult)
    Handler->>Broker: Publish RequirementGenerationCompleted
    Broker->>API: Deliver Completed Event

Business Intent

The goal of the Requirements Generation feature is to ensure that functional, structural, and safety requirements are derived effectively from the actual system context or source files. It guarantees that:

• Every requirement is uniquely identifiable.
• Duplicate descriptions are purged automatically to maintain a clean requirement specification.
• A consistent safety level (e.g., QM) and status (e.g., Reviewed) are enforced out-of-the-box.
• When generating multiple requirements from a single block, it prepares one “Update” to mutate the existing block, and multiple “Creates” for its siblings, providing a seamless data-entry transition in the UI.