Skill CD CI Ddd Score Validation

1. Purpose

Audit any codebase to verify that Clean Architecture / DDD layer ordering is respected:

Domain <- Application <- UseCases <- Infrastructure
Domain <- Application <- Interface

The agent (LLM) is the primary auditor. Companion shell scripts in tools/ extract raw data (file lists, imports, dependency graph). The agent interprets this data, applies subjective criteria, computes a score, and writes the final YAML report.

2. When to Use

• New project onboarding: assess architecture health.
• Pre-refactoring: identify violations before restructuring.
• CI/CD gate: generate a score report on every PR.
• Code review: validate that changes respect layer boundaries.

3. Workflow (7 Steps)

Step 1: Discover Source Root

Invoke: bash tools/discover-source.sh <project_path> Purpose: find the main source folder (src/, source/, app/, lib/, etc.).

Step 2: Discover Layers

Invoke: bash tools/discover-layers.sh <source_root> Purpose: map which layer folders exist inside the source root.

Step 3: Classify Files

Invoke: bash tools/classify-files.sh <source_root> Purpose: classify every file by layer, language, and responsibility type.

Step 4: Extract Imports

Invoke: bash tools/extract-imports.sh <classified_files_yaml> Purpose: build a dependency graph (imports per file).

Step 5: Validate Hard Rules

Invoke:

• bash tools/validate-dependencies.sh <imports_yaml> <layers_yaml>
• bash tools/validate-naming.sh <classified_files_yaml>
• bash tools/detect-logic-leaks.sh <classified_files_yaml>

Purpose: obtain objective, machine-detectable violations.

Step 6: Apply Subjective Audit (LLM)

The agent reads key files directly (using its own Read/Glob capabilities) to evaluate:

• Are DTOs flat in the Application layer?
• Is there one responsibility per file?
• Are LAYER header comments present?
• Are repository interfaces defined in Domain?
• Does the Dependency Inversion principle hold in practice?

The agent combines:

• Hard violations (weight ~70%)
• Subjective findings (weight ~30%)
• Scoring matrix from references/scoring-matrix.md

Step 7: Generate Final YAML Report

The agent writes ddd-score-report.yml with exact schema defined in references/output-schema.yml.

4. Tool Invocation Guide

All scripts are 100% bash, read from the filesystem, and write YAML to stdout.

4.1 discover-source.sh

bash tools/discover-source.sh /path/to/project

Output:

project_path: /path/to/project
source_root: /path/to/project/src
candidates:
  - path: src
    file_count: 142
    languages:
      - typescript
      - javascript
  - path: app
    file_count: 12
    languages:
      - typescript
selected: src
selected_reason: highest file count (142)
frameworks_detected:
  - qwik
  - prisma
  - authjs

4.2 discover-layers.sh

bash tools/discover-layers.sh /path/to/project/src

Output:

source_root: /path/to/project/src
layers_detected:
  - name: Domain
    path: src/domain
    file_count: 12
  - name: Application
    path: src/application
    file_count: 8
  - name: Infrastructure
    path: src/infrastructure
    file_count: 6
  - name: Presentation
    path: src/@presentation
    canonical_role: Interface
    file_count: 991

Note: presentation/, ui/, views/ are canonicalized as Presentation with canonical_role: Interface. @services/ is not a standard layer; it defaults to Infrastructure.

4.3 classify-files.sh

bash tools/classify-files.sh /path/to/project/src

Output:

source_root: /path/to/project/src
files:
  - path: src/domain/entities/user.entity.ts
    layer: Domain
    language: typescript
    type: entity
    basename: user.entity.ts
  - path: src/application/use-cases/create-user.use-case.ts
    layer: Application
    language: typescript
    type: use-case
    basename: create-user.use-case.ts

4.4 extract-imports.sh

bash tools/extract-imports.sh /path/to/classified-files.yml

Output:

imports:
  - source_file: src/application/use-cases/create-user.use-case.ts
    imported_path: ../../domain/entities/user.entity
    is_relative: true
    is_type_only: false
    import_line: "import { User } from '../../domain/entities/user.entity';"
    language: typescript
  - source_file: src/presentation/store/billing.ts
    imported_path: ../../domain/entities/BillingInformation
    is_relative: true
    is_type_only: true
    import_line: "import type { BillingInformationDTO } from '../../domain/entities/BillingInformation';"
    language: typescript

Note: is_type_only: true indicates a compile-time-only import. The agent should treat these as lower severity than concrete imports.

4.5 validate-dependencies.sh

bash tools/validate-dependencies.sh /path/to/imports.yml /path/to/layers.yml

Output:

violations:
  - file: src/domain/entities/user.entity.ts
    rule: DOMAIN_NO_EXTERNAL_DEPS
    severity: critical
    message: Domain entity imports from Infrastructure
    suggestion: "Move 'import { PrismaClient }' from src/domain/entities/user.entity.ts to an Infrastructure repository implementation."
    imported_layer: Infrastructure
    source_layer: Domain

4.6 validate-naming.sh

bash tools/validate-naming.sh /path/to/classified-files.yml

Output:

naming_issues:
  - file: src/application/createUser.ts
    layer: Application
    expected_pattern: "*.use-case.ts"
    actual: createUser.ts
    severity: medium
    suggestion: "Rename to create-user.use-case.ts"

4.7 detect-logic-leaks.sh

bash tools/detect-logic-leaks.sh /path/to/classified-files.yml

Output:

logic_leaks:
  - file: src/interface/controllers/user.controller.ts
    layer: Interface
    keyword_found: calculateTax
    context_line: "const tax = calculateTax(amount, rate);"
    severity: high
    suggestion: "Move business logic to a Domain Service or Application Use Case."

4.8 aggregate-raw-data.sh (Convenience)

bash tools/aggregate-raw-data.sh /path/to/project

Runs steps 1-7 in sequence and emits a combined YAML document.

5. Hard Rules Reference

Rules validated automatically by scripts or enforced by the agent:

6. Subjective Criteria (Agent Evaluation)

The agent must read files and judge:

• DTO quality: Are Application DTOs flat objects without ORM annotations?
• Header comments: Does every file start with a LAYER: documentation comment?
• Business logic placement: Is there business logic hidden in controllers, DTOs, or mappers?
• Dependency inversion: Are dependencies injected via constructor (or simple composition root) rather than instantiated directly?
• Anemic domain model: Do entities contain behavior, or are they just data bags?
• Use case purity: Do use cases only orchestrate, or do they contain business rules?

7. Scoring Matrix

Score is computed by the agent using references/scoring-matrix.md.

Quick reference:

8. YAML Output Specification

The agent must generate a file matching references/output-schema.yml.

Mandatory top-level keys:

• project
• score
• max_score
• summary (human-readable paragraph)
• layers_detected[]
• violations[] (each with suggestion for concrete auto-fix)
• subjective_notes[]
• todo_desirable[] (each with priority: P0/P1/P2, layer, task)

9. Todo Templates

For inspiration generating todo_desirable, see references/todo-templates.md.

10. CI/CD Usage

# 1. Extract raw data
bash tools/aggregate-raw-data.sh . > /tmp/raw-audit.yml

# 2. Agent reads raw-audit.yml + source files
# 3. Agent writes ddd-score-report.yml

In GitHub Actions, steps 2-3 are performed by the agent/LLM step after the raw data is extracted.

Build specs that engineering can execute without asking "what did you mean?"