From rails-agent-skills
Enforces Ruby on Rails service object conventions: .call pattern, YARD docs on self.call/public methods, module READMEs, {success:, response:} contract, orchestration, transactions, error handling. Use for creating/refactoring app/services.
npx claudepluginhub igmarin/rails-agent-skills --plugin rails-agent-skillsThis skill uses the workspace's default tool permissions.
| Aspect | Rule |
Guides RSpec testing for service objects, API clients, orchestrators, and business logic in spec/services/, using instance_double, FactoryBot, aggregate_failures, change matchers, and travel_to.
Provides Ruby coding conventions and best practices for naming, service objects, RSpec testing, error handling, AWS SDK usage, logging, and security patterns.
Guides Rails code with rubocop-rails-omakase conventions on Turbo Streams, controllers, concerns, services, modern Ruby style, enums, scopes, error handling, and Minitest testing.
Share bugs, ideas, or general feedback.
| Aspect | Rule |
|---|---|
| Entry point | def self.call(...) → new(...).call |
| Validation | Validate inputs at top of call; return error hash if invalid |
| Error handling | rescue → log + error hash; never re-raise to caller |
| Transactions | Only wrap multi-step DB operations that must be atomic |
call length | ≤20 lines; extract sub-services if longer |
| Scope | Return data only (no HTTP); single responsibility per service |
| SQL | sanitize_sql for any dynamic queries |
| Shared logic | Extract validators to class-only services (Pattern 3) |
Required artifacts (file path, README, YARD, pragma, error log shape) are enumerated once in Output Style.
EVERY service object MUST have its test written and validated BEFORE implementation.
1. Write the spec for .call (with contexts for success, error, edge cases)
2. Run the spec — verify it fails because the service does not exist yet
3. ONLY THEN write the service implementation
See rspec-best-practices for the full gate cycle.
.call MUST return EXACTLY one of these two shapes — no other top-level keys, never a Boolean, raw model, or ActiveRecord object:
{ success: true, response: { <domain_key>: <value>, ... } }
{ success: false, response: { error: { message: 'human-readable reason' } } }
| Signal in the task | Pattern |
|---|---|
| Orchestrates multiple steps, needs instance state | Pattern 1: .call → new.call |
| Processes a collection with per-item error handling | Pattern 2: Batch processing |
| Stateless helper, validator, or utility — no instance state needed | Pattern 3: Class-only (static methods) |
| Coordinates multiple sub-services | Pattern 4: Orchestrator delegation |
.call Pattern (with delegation, transaction, YARD)module AnimalTransfers
class TransferService
TRANSFER_FAILED = 'Transfer could not be completed'
# @param params [Hash] :source_shelter_id, :target_shelter_id, :tag_number
# @return [Hash] { success: Boolean, response: Hash }
def self.call(params)
new(params).call
end
def initialize(params)
@source_shelter_id = params[:source_shelter_id]
@target_shelter_id = params[:target_shelter_id]
@tag_number = params[:tag_number]
end
def call
source = ShelterValidator.validate_source_shelter!(@source_shelter_id)
target = ShelterValidator.validate_target_shelter!(@target_shelter_id)
result = ActiveRecord::Base.transaction do
source.decrement!(:animal_count)
target.increment!(:animal_count)
TransferLog.create!(source:, target:, tag_number: @tag_number)
end
{ success: true, response: { transfer: result } }
rescue ActiveRecord::RecordInvalid => e
Rails.logger.error("Validation Error: #{e.message}")
Rails.logger.error(e.backtrace.join("\n"))
{ success: false, response: { error: { message: e.message } } }
rescue StandardError => e
Rails.logger.error("Processing Error: #{e.message}")
Rails.logger.error(e.backtrace.join("\n"))
{ success: false, response: { error: { message: TRANSFER_FAILED } } }
end
end
end
# Batch — each rescue block logs; outer rescue returns { success: false }
def call
return { success: false, response: { error: { message: 'Items list cannot be empty' } } } if @items.blank?
results = @items.each_with_object({ successful: [], failed: [] }) do |item, acc|
validate_item!(item)
process_item(item)
acc[:successful] << item[:sku]
rescue ActiveRecord::RecordNotFound => e
Rails.logger.error("Item not found: #{e.message}")
acc[:failed] << { sku: item[:sku], error: e.message }
rescue StandardError => e
Rails.logger.error("Unexpected item error: #{e.message}")
Rails.logger.error(e.backtrace.join("\n"))
acc[:failed] << { sku: item[:sku], error: e.message }
end
{ success: true, response: results }
rescue StandardError => e
Rails.logger.error("Service failed: #{e.message}")
Rails.logger.error(e.backtrace.join("\n"))
{ success: false, response: { error: { message: PROCESSING_FAILED } } }
end
When no instance state is needed — use ONLY class methods, no initialize, no instance variables. Validators and stateless helpers should always use this pattern:
class PackageValidator
MAX_WEIGHT_KG = 30
MAX_LENGTH_CM = 150
# @param dimensions [Hash] :weight_kg, :length_cm, :width_cm, :height_cm
# @return [nil, String] nil if valid, error message otherwise
def self.validate(dimensions)
return 'Weight exceeds limit' if dimensions[:weight_kg] > MAX_WEIGHT_KG
return 'Length exceeds limit' if dimensions[:length_cm] > MAX_LENGTH_CM
nil
end
def self.within_limits?(dimensions)
validate(dimensions).nil?
end
end
call)Sub-services handle their OWN rescue and return { success: false, response: { error: { message: ... } } } on failure. The orchestrator propagates early returns only — no rescue block needed:
# RULE: ≤20 lines in call — if longer, extract another sub-service
def call
user_result = UserCreationService.call(@params)
return user_result unless user_result[:success]
workspace_result = WorkspaceSetupService.call(user_result[:response])
return workspace_result unless workspace_result[:success]
BillingService.call(workspace_result[:response])
NotificationService.call(user_result[:response])
{ success: true, response: { user: user_result[:response] } }
end
Every service-object task produces these artifacts:
app/services/<module_name>/<service_name>.rb — pragma on line 1, class wrapped in a module matching the directory name.self.call — @param for every argument, @return [Hash], plus @raise for any exception class that can escape (including those rescued internally elsewhere). The self.call wrapper is documented separately from #call, even when it just delegates.@param / @return / @raise discipline. See yard-documentation.UPPER_SNAKE_CASE constants at the top of the class (e.g. TRANSFER_FAILED = 'Transfer could not be completed'), never inline inside a rescue.app/services/<module_name>/README.md — copy the shape from assets/module_readme_template.md. Required even for single-service modules.spec/services/<module_name>/<service_name>_spec.rb written and failing BEFORE the implementation (see HARD-GATE).For class-only services (Pattern 3), the rules apply to the public class methods being documented; if the class returns a non-service shape (e.g. validators returning nil / error string), document that explicitly in YARD and the README.
| Skill | When to chain |
|---|---|
| yard-documentation | Writing/reviewing inline docs |
| ruby-api-client-integration | External API integrations |
| strategy-factory-null-calculator | Variant-based calculators |
| rspec-service-testing | Testing service objects |
| rspec-best-practices | General RSpec structure |
| rails-architecture-review | Architecture review involving service extraction |