🧪 Skills

FOSMVVM Fluent DataModel Generator

Generate Fluent DataModels with migrations and tests for FOSMVVM server-side persistence, including relationships and system-assigned fields, based on existi...

v2.0.6
❤️ 0
⬇️ 405
👁 1
Share

Description


name: fosmvvm-fluent-datamodel-generator description: Generate Fluent DataModels for FOSMVVM server-side persistence. Scaffolds models, migrations, and tests for database-backed entities. homepage: https://github.com/foscomputerservices/FOSUtilities metadata: {"clawdbot": {"emoji": "🗄️", "os": ["darwin", "linux"]}}

FOSMVVM Fluent DataModel Generator

Generate Fluent DataModels for server-side persistence following FOSMVVM architecture.

Dependency: This skill uses fosmvvm-fields-generator for the Fields layer (protocol, messages, YAML). Run that skill first for form-backed models.

Scope Guard

This skill is specifically for Fluent persistence layer (typically in Vapor server apps).

STOP and ask the user if:

  • The project doesn't use Fluent
  • The target is iOS-only with CoreData, SwiftData, or Realm
  • The user mentions a non-Fluent ORM or persistence layer
  • You're unsure whether Fluent is the persistence layer

Check for Fluent indicators:

  • Package.swift imports fluent, fluent-postgres-driver, fluent-sqlite-driver, etc.
  • Existing models use @ID, @Field, @Parent, @Children, @Siblings property wrappers
  • A Migrations/ directory exists with Fluent migration patterns
  • Imports include FluentKit or Fluent

If Fluent isn't present, inform the user: "This skill generates Fluent DataModels for server-side persistence. Your project doesn't appear to use Fluent. How would you like to proceed?"


When to Use This Skill

  • User asks to create a new model/entity/table
  • User wants to add a database-backed type (Users, Ideas, Documents, etc.)
  • User mentions needing CRUD operations for a new concept
  • Creating the persistence layer for a new entity

Architecture Context

In FOSMVVM, the Model is the center - the source of truth that reads and writes flow through.

See FOSMVVMArchitecture.md | OpenClaw reference for full context.

DataModel in the Architecture

                    ┌─────────────────────────────────────┐
                    │         Fluent DataModel            │
                    │    (implements Model + Fields)      │
                    │                                     │
                    │  • All fields (system + user)       │
                    │  • Relationships (@Parent, etc.)    │
                    │  • Timestamps, audit fields         │
                    │  • Persistence logic                │
                    └──────────────┬──────────────────────┘
                                   │
              ┌────────────────────┼────────────────────────┐
              │                    │                        │
              ▼                    ▼                        ▼
    ┌─────────────────┐  ┌─────────────────┐    ┌─────────────────┐
    │ ViewModelFactory│  │  CreateRequest  │    │  UpdateRequest  │
    │   (projector)   │  │   RequestBody   │    │   RequestBody   │
    │                 │  │                 │    │                 │
    │ → ViewModel     │  │ → persists to   │    │ → updates       │
    │   (projection)  │  │   DataModel     │    │   DataModel     │
    └─────────────────┘  └─────────────────┘    └─────────────────┘

Fields vs DataModels

Fields protocol = Form input (user-editable subset)

  • What users type into forms
  • Validation, labels, placeholders
  • NO relationships, NO system-assigned fields

DataModel = Complete entity (Fluent implementation)

  • All fields including system-assigned (createdBy, timestamps)
  • All relationships (@Parent, @Siblings, @Children)
  • Fluent property wrappers, migrations, seeds

Not all entities need Fields:

  • Session: system auth, no user form → DataModel-only
  • Audit records: system-generated → DataModel-only
  • Junction tables: pure storage → DataModel-only

File Structure

Each form-backed model requires files across multiple targets:

── fosmvvm-fields-generator ──────────────────────────────────
{ViewModelsTarget}/                  (shared protocol layer)
  FieldModels/
    {Model}Fields.swift              ← Protocol + Enum + Validation
    {Model}FieldsMessages.swift      ← Localization message struct

{ResourcesPath}/                     (localization resources)
  FieldModels/
    {Model}FieldsMessages.yml        ← YAML localization strings

── fosmvvm-fluent-datamodel-generator (this skill) ───────────
{WebServerTarget}/                   (server implementation)
  DataModels/
    {Model}.swift                    ← Fluent model (implements protocol)
  Migrations/
    {Model}+Schema.swift             ← Table creation migration
    {Model}+Seed.swift               ← Seed data migration

Tests/
  {ViewModelsTarget}Tests/
    FieldModels/
      {Model}FieldsTests.swift       ← Unit tests

database.swift                       ← Register migrations

How to Use This Skill

Invocation: /fosmvvm-fluent-datamodel-generator

Prerequisites:

  • Model structure understood from conversation context
  • Fields protocol exists (if form-backed model) via fosmvvm-fields-generator
  • Relationships and system-assigned fields identified
  • Fluent confirmed as the persistence layer

Workflow integration: This skill is used for server-side persistence with Fluent. For form-backed models, run fosmvvm-fields-generator first to create the Fields protocol. The skill references conversation context automatically—no file paths or Q&A needed.

Pattern Implementation

This skill references conversation context to determine DataModel structure:

Model Type Detection

From conversation context, the skill identifies:

  • Entity purpose (user data, system records, audit logs, junction table)
  • User input involvement (form-backed vs system-generated)
  • Fields protocol requirement (if user edits this data)

Relationship Analysis

From requirements already in context:

  • One-to-many relationships (@Parent in DataModel, not in Fields)
  • Many-to-many relationships (Junction table + @Siblings, NOT UUID arrays)
  • Relationship naming (self-documenting names, not vague references)

Field Classification

Based on data source:

  • User-editable fields (from Fields protocol)
  • System-assigned fields (createdBy, timestamps, status - DataModel only)
  • Computed relationships (@Parent, @Children, @Siblings)

File Generation Order

If form-backed model (Fields protocol exists):

  1. Fields layer already created via fosmvvm-fields-generator
  2. DataModel implementation referencing Fields
  3. Schema migration
  4. Seed data migration
  5. Tests
  6. Migration registration

If system-only model (no Fields):

  1. DataModel struct
  2. Schema migration
  3. Seed data migration (if needed)
  4. Tests
  5. Migration registration

Design Validation

Before generating, the skill validates:

  1. Form requirement - System-generated entities skip Fields
  2. Relationship patterns - Junction tables for many-to-many, @Parent for foreign keys
  3. Naming clarity - Relationships have self-documenting names
  4. Field separation - User fields in protocol, system fields in DataModel only

Context Sources

Skill references information from:

  • Prior conversation: Model requirements, relationships discussed
  • Fields protocol: If Claude has read Fields protocol into context or just created it
  • Database schema: From codebase analysis of existing models
  • Migration patterns: From existing migrations in project

File Templates

See reference.md for complete file templates with all patterns.


Key Patterns

Fluent DataModel

import FluentKit
import FOSFoundation
import FOSMVVM
import FOSMVVMVapor
import Foundation

final class {Model}: DataModel, {Model}Fields, Hashable, @unchecked Sendable {
    static let schema = "{models}"  // snake_case plural

    @ID(key: .id) var id: ModelIdType?

    // Fields from protocol
    @Field(key: "field_name") var fieldName: FieldType

    // Validation messages
    let {model}ValidationMessages: {Model}FieldsMessages

    // Timestamps
    @Timestamp(key: "created_at", on: .create) var createdAt: Date?
    @Timestamp(key: "updated_at", on: .update) var updatedAt: Date?

    // CRITICAL: Initialize validationMessages FIRST
    init() {
        self.{model}ValidationMessages = .init()
    }

    init(id: ModelIdType? = nil, fieldName: FieldType) {
        self.{model}ValidationMessages = .init()  // FIRST!
        self.id = id
        self.fieldName = fieldName
    }
}

Relationships (Associated Types Pattern)

PRINCIPLE: Existential types (any Protocol) are a code smell. Always ask "Is there any other way?" before using them.

For required relationships, use associated types in the protocol:

public protocol IdeaFields: ValidatableModel, Codable, Sendable {
    associatedtype User: UserFields

    var createdBy: User { get set }
}

In the Fluent model, @Parent directly satisfies the protocol:

final class Idea: DataModel, IdeaFields, Hashable, @unchecked Sendable {
    @Parent(key: "created_by") var createdBy: User
    // No computed property needed - @Parent satisfies the associated type directly
}

In schema: .field("created_by", .uuid, .required, .references(User.schema, "id", onDelete: .cascade))

When to use each pattern:

  • Associated type (associatedtype User: UserFields): Required relationships
  • Optional associated type: Not supported - use ModelIdType? for optional FKs
  • Plain ModelIdType: Optional FKs, external system references

Migrations

  • Schema migration named: "{Model.schema}-initial"
  • Seed migration named: "{Model.schema}-seed"
  • Seed is environment-aware (debug, test, release)
  • Seed is idempotent: guard count() == 0

Raw SQL in Migrations (PostgreSQL Features)

For PostgreSQL-specific features (tsvector, LTREE, etc.), use SQLKit:

import Fluent
import SQLKit  // Required for raw SQL

// In prepare():
guard let sql = database as? any SQLDatabase else { return }

let schema = Model.schema
try await sql.raw(SQLQueryString("ALTER TABLE \(unsafeRaw: schema) ADD COLUMN search_vector tsvector")).run()

Key points:

  • Import SQLKit (not just Fluent)
  • Cast database: database as? any SQLDatabase
  • Use SQLQueryString with \(unsafeRaw:) for identifiers
  • These columns are database-only (not in protocol or Fluent model)

Tests

  • Use @Suite annotation with descriptive name
  • Conform to LocalizableTestCase
  • Test all form fields
  • Test validation with @Test(arguments:)
  • Create private test struct implementing the Fields protocol

Test structs with associated types:

private struct TestIdea: IdeaFields {
    typealias User = TestUser  // Satisfy the associated type

    var id: ModelIdType?
    var createdBy: TestUser    // Concrete type, not existential
}

private struct TestUser: UserFields {
    var id: ModelIdType? = .init()
    var firstName: String = "Test"
    // ... other required fields with defaults
}

Naming Conventions

Concept Convention Example
Model class PascalCase singular User, Idea
Table name snake_case plural users, ideas
Field keys snake_case created_at, user_id
Enum cases camelCase searchLanguage, inProgress
Enum raw values snake_case "search_language", "in_progress"
Protocol {Model}Fields UserFields, IdeaFields
Messages struct {Model}FieldsMessages UserFieldsMessages

Common Field Types

Swift Type Fluent Type Database
String .string VARCHAR/TEXT
Int .int INTEGER
Bool .bool BOOLEAN
Date .datetime TIMESTAMPTZ
UUID .uuid UUID
[UUID] .array(of: .uuid) UUID[]
Custom Enum .string VARCHAR (stored as raw value)
JSONB .json JSONB

See Also


Version History

Version Date Changes
1.0 2025-12-23 Initial skill based on SystemConfig pattern
1.1 2025-12-23 Added relationship patterns (@Parent), initialization order, imports list
1.2 2025-12-23 Associated types for relationships (not existentials), raw SQL patterns, test struct patterns
1.3 2025-12-24 Factored out Fields layer to fields-generator skill
2.0 2025-12-26 Renamed to fosmvvm-fluent-datamodel-generator, added Scope Guard, generalized from Kairos-specific to FOSMVVM patterns, added architecture context
2.1 2026-01-24 Update to context-aware approach (remove file-parsing/Q&A). Skill references conversation context instead of asking questions or accepting file paths.

Reviews (0)

Sign in to write a review.

No reviews yet. Be the first to review!

Comments (0)

Sign in to join the discussion.

No comments yet. Be the first to share your thoughts!

Compatible Platforms

Pricing

Free

Related Configs