Agent skill
services-layer
Service layer patterns with defineErrors, namespace exports, and Result types. Use when the user says "create a service", "service layer", or when creating new services, defining domain-specific errors, or understanding the service architecture.
Install this agent skill to your Project
npx add-skill https://github.com/EpicenterHQ/epicenter/tree/main/.agents/skills/services-layer
Metadata
Additional technical details for this skill
- author
- epicenter
- version
- 2.0
SKILL.md
Services Layer Patterns
This skill documents how to implement services in the Whispering architecture. Services are pure, isolated business logic with no UI dependencies that return Result<T, E> types for error handling.
Related Skills: See
error-handlingfor trySync/tryAsync patterns. Seedefine-errorsfor error variant factories. Seequery-layerfor consuming services with TanStack Query.
When to Apply This Skill
Use this pattern when you need to:
- Create a new service with domain-specific error handling
- Add error types with structured context (like HTTP status codes)
- Understand how services are organized and exported
- Implement platform-specific service variants (desktop vs web)
Core Architecture
Services follow a three-layer architecture: Service → Query → UI
┌─────────────┐ ┌─────────────┐ ┌──────────────┐
│ UI │ --> │ RPC/Query │ --> │ Services │
│ Components │ │ Layer │ │ (Pure) │
└─────────────┘ └─────────────┘ └──────────────┘
Services are:
- Pure: Accept explicit parameters, no hidden dependencies
- Isolated: No knowledge of UI state, settings, or reactive stores
- Testable: Easy to unit test with mock parameters
- Consistent: All return
Result<T, E>types for uniform error handling
Creating Errors with defineErrors
Every service defines domain-specific errors using defineErrors from wellcrafted. Errors are grouped into a namespace object where each key becomes a variant.
import { defineErrors, type InferError, type InferErrors, extractErrorMessage } from 'wellcrafted/error';
import { Err, Ok, type Result, tryAsync, trySync } from 'wellcrafted/result';
// Namespace-style error definition — name describes the domain
const CompletionError = defineErrors({
ConnectionFailed: ({ cause }: { cause: unknown }) => ({
message: `Connection failed: ${extractErrorMessage(cause)}`,
cause,
}),
EmptyResponse: ({ providerLabel }: { providerLabel: string }) => ({
message: `${providerLabel} API returned an empty response`,
providerLabel,
}),
MissingParam: ({ param }: { param: string }) => ({
message: `${param} is required`,
param,
}),
});
// Type derivation — shadow the const with a type of the same name
type CompletionError = InferErrors<typeof CompletionError>;
type ConnectionFailedError = InferError<typeof CompletionError.ConnectionFailed>;
// Call sites — each variant returns Err<...> directly
return CompletionError.ConnectionFailed({ cause: error });
return CompletionError.EmptyResponse({ providerLabel: 'OpenAI' });
return CompletionError.MissingParam({ param: 'apiKey' });
How defineErrors Works
defineErrors({ ... }) takes an object of factory functions and returns a namespace object. Each key becomes a variant:
nameis auto-stamped from the key (e.g., keyNotFound→error.name === 'NotFound')- The factory function IS the message generator — it returns
{ message, ...fields } - Each variant returns
Err<...>directly — no separateFooErrconstructor needed - Types use
InferError/InferErrors— notReturnType
// No-input variant (static message)
const RecorderError = defineErrors({
Busy: () => ({
message: 'A recording is already in progress',
}),
});
// Usage — no arguments needed
return RecorderError.Busy();
// Variant with derived fields — constructor extracts from raw input
const HttpError = defineErrors({
Response: ({ response, body }: { response: { status: number }; body: unknown }) => ({
message: `HTTP ${response.status}: ${extractErrorMessage(body)}`,
status: response.status,
body,
}),
});
// Usage — pass raw objects, constructor derives fields
return HttpError.Response({ response, body: await response.json() });
// error.message → "HTTP 401: Unauthorized"
// error.status → 401 (derived from response, flat on the object)
// error.name → "Response"
Error Type Examples from the Codebase
// Static message, no input needed
const RecorderError = defineErrors({
Busy: () => ({
message: 'A recording is already in progress',
}),
});
RecorderError.Busy()
// Multiple related errors in a single namespace
const HttpError = defineErrors({
Connection: ({ cause }: { cause: unknown }) => ({
message: `Failed to connect to the server: ${extractErrorMessage(cause)}`,
cause,
}),
Response: ({ response, body }: { response: { status: number }; body: unknown }) => ({
message: `HTTP ${response.status}: ${extractErrorMessage(body)}`,
status: response.status,
body,
}),
Parse: ({ cause }: { cause: unknown }) => ({
message: `Failed to parse response body: ${extractErrorMessage(cause)}`,
cause,
}),
});
// Union type for the whole namespace
type HttpError = InferErrors<typeof HttpError>;
// Individual variant type
type ConnectionError = InferError<typeof HttpError.Connection>;
Key Rules
- Services never import settings - Pass configuration as parameters
- Services never import UI code - No toasts, no notifications, no WhisperingError
- Always return Result types - Never throw errors
- Use trySync/tryAsync - See the error-handling skill for details
- Export factory + Live instance - Factory for testing, Live for production
- Use defineErrors namespaces - Group related errors under a single namespace
- Derive types with InferError/InferErrors - Not
ReturnType - Variant names describe the failure mode - Never use generic names like
Service,Error, orFailed. The namespace provides domain context (RecorderError), so the variant must say what went wrong (AlreadyRecording,InitFailed,StreamAcquisition).RecorderError.Serviceis meaningless —RecorderError.AlreadyRecordingtells you exactly what happened. - Split discriminated union inputs - Each variant gets its own name and shape. If the constructor branches on its inputs (if/switch/ternary) to decide the message, each branch should be its own variant
- Transform cause in the constructor, not the call site - Accept
cause: unknownand callextractErrorMessage(cause)inside the factory's message template. Call sites pass the raw error:{ cause: error }. This centralizes message extraction where the message is composed and keeps call sites clean.
References
Load these on demand based on what you're working on:
-
If working with error variant anti-patterns (discriminated union inputs, branching constructors), read references/error-anti-patterns.md
-
If working with service implementation details (factory patterns, recorder service examples), read references/service-implementation-pattern.md
-
If working with service organization and platform variants (namespace exports, desktop vs web services), read references/service-organization-platforms.md
-
If working with error message authoring (user-friendly/actionable message design), read references/error-message-best-practices.md
-
See
apps/whispering/src/lib/services/README.mdfor architecture details -
See the
query-layerskill for how services are consumed -
See the
error-handlingskill for trySync/tryAsync patterns
Recommended Agent Skills
Expand your agent's capabilities with these related and highly-rated skills.
svelte
Svelte 5 patterns including runes ($state, $derived, $props), TanStack Query, SvelteMap reactive state, shadcn-svelte components, and component composition. Use when the user mentions .svelte files, Svelte components, or when using TanStack Query, fromTable/fromKv, or shadcn-svelte UI.
autumn
Integrate Autumn billing—define features/plans in autumn.config.ts, use autumn-js SDK for credit checks/tracking, manage the atmn CLI for push/pull. Use when working on billing, pricing, credits, plan gating, or metered usage.
handoff-prompt
Draft a self-contained implementation prompt that an agent can execute with zero prior context. Use when the user says "draft a prompt", "write a handoff", "make a prompt I can copy-paste", "create a delegation brief", or wants to hand off a task to another agent, tool, or conversation.
typebox
TypeBox and TypeMap patterns for runtime schema validation and JSON Schema generation. Use when the user mentions TypeBox, TypeMap, Standard Schema, or when working with runtime type validation, JSON Schema, or schema-based validation.
factory-function-composition
Apply factory function patterns to compose clients and services with proper separation of concerns. Use when creating functions that depend on external clients, wrapping resources with domain-specific methods, or refactoring code that mixes client/service/method options together.
progress-summary
This skill should be used when the user asks questions like "can you summarize", "what happened", "what did we do", "what's the situation", "where are we at", "explain what's going on", "give me an overview", "what's been done", "tell me about this", "walk me through what happened", or any question asking to understand the current state of work or changes. Provides conversational, PR-style summaries with visual diagrams.
Didn't find tool you were looking for?