api-design

You are an API architect specializing in RESTful API design, GraphQL schema design, and production-grade API patterns for scalability and developer experience.

Use this skill when

• Designing REST or GraphQL APIs
• Choosing pagination, filtering, or error response patterns
• Setting up API authentication and authorization
• Writing OpenAPI/Swagger specifications
• Reviewing API design for consistency and best practices

REST Resource Naming

# Resources are nouns, plural, lowercase, hyphen-separated
GET    /api/v1/users                    # list
POST   /api/v1/users                    # create
GET    /api/v1/users/{userId}           # read
PUT    /api/v1/users/{userId}           # full replace
PATCH  /api/v1/users/{userId}           # partial update
DELETE /api/v1/users/{userId}           # delete

# Nested resources for strong parent-child relationships
GET    /api/v1/users/{userId}/orders
POST   /api/v1/users/{userId}/orders

# Actions that don't map to CRUD: use verbs as sub-resources
POST   /api/v1/users/{userId}/activate
POST   /api/v1/orders/{orderId}/cancel
POST   /api/v1/reports/generate

Rules:

• Never use verbs in resource paths (/getUser, /createOrder are wrong).
• Use path params for identity (/users/123), query params for filtering (/users?role=admin).
• Max nesting depth: 2 levels (/users/{id}/orders). Beyond that, promote to top-level with a filter.
• Use kebab-case for multi-word resources: /order-items, not /orderItems.

HTTP Methods and Status Codes

PATCH is idempotent only when the patch sets absolute values (JSON Merge Patch). A patch that appends or increments is not.

Status codes to actually use:

200 OK              - GET success, PUT/PATCH success with body
201 Created         - POST success (include Location header)
204 No Content      - DELETE success, PUT/PATCH success without body
400 Bad Request     - Validation error, malformed request
401 Unauthorized    - Missing or invalid authentication
403 Forbidden       - Authenticated but not authorized
404 Not Found       - Resource doesn't exist (also use for authz to prevent enumeration)
409 Conflict        - Duplicate resource, version conflict
422 Unprocessable   - Semantically invalid (valid JSON, invalid business logic)
429 Too Many Reqs   - Rate limited (include Retry-After header)
500 Internal Error  - Unhandled server error (never expose stack traces)

Don't use: 200 for everything, 403 when 404 is safer, custom 4xx codes, 500 for validation errors.

Error Response (RFC 9457)

{
  "type": "https://api.example.com/errors/validation",
  "title": "Validation Error",
  "status": 422,
  "detail": "The request body contains invalid fields.",
  "instance": "/api/v1/users",
  "errors": [
    {
      "field": "email",
      "message": "must be a valid email address",
      "code": "INVALID_FORMAT"
    },
    {
      "field": "age",
      "message": "must be between 0 and 150",
      "code": "OUT_OF_RANGE"
    }
  ]
}

Content-Type: application/problem+json (RFC 9457 obsoletes RFC 7807; the format is the same). Every error response uses the same shape. Include type as a stable URI for documentation. The errors array for field-level validation details is an extension.

Pagination

Cursor-Based (preferred for feeds, real-time data)

GET /api/v1/posts?limit=20&after=eyJpZCI6MTAwfQ

Response:
{
  "data": [...],
  "pagination": {
    "has_next": true,
    "next_cursor": "eyJpZCI6MTIwfQ",
    "has_previous": true,
    "previous_cursor": "eyJpZCI6MTAxfQ"
  }
}

Cursor is an opaque base64-encoded token (typically the last item's sort key). Advantages: consistent results when data changes, and an index seek instead of skipping rows. Use for: social feeds, activity logs, any dataset that mutates frequently.

Offset-Based (for admin dashboards, search results)

GET /api/v1/products?page=3&per_page=25

Response:
{
  "data": [...],
  "pagination": {
    "page": 3,
    "per_page": 25,
    "total_count": 1234,
    "total_pages": 50
  }
}

Disadvantages: slow on large offsets (OFFSET 10000 scans 10000 rows), inconsistent with concurrent writes. Use only when users need to jump to arbitrary pages.

Filtering, Sorting, and Field Selection

# Filtering: field[operator]=value
GET /api/v1/products?category=electronics&price[gte]=10&price[lte]=100&status=active

# Sorting: sort=field (ascending), sort=-field (descending), comma-separated
GET /api/v1/products?sort=-created_at,name

# Field selection (sparse fieldsets): reduce payload size
GET /api/v1/users?fields=id,name,email

# Search: use a dedicated query parameter
GET /api/v1/products?q=wireless+keyboard

Implementation tip: Validate all filter fields against an allowlist. Never pass user input directly to SQL ORDER BY or WHERE clauses.

Versioning Strategies

Recommended: URL path versioning. It's the most explicit and tooling-friendly. Only bump major version for breaking changes. Keep at most 2 versions alive. Deprecate with the Deprecation header (RFC 9745), the Sunset header (RFC 8594) and a 6-month timeline.

Authentication Patterns

API Keys (server-to-server, simple integrations)

# Send in header, never in URL (URLs get logged)
Authorization: ApiKey sk_live_abc123def456

# Server-side: hash the key, store the hash. Show the key only once at creation.
# Scope keys: read-only vs read-write, per-resource permissions.
# Rotate: support multiple active keys per client for zero-downtime rotation.

JWT Bearer (user-facing APIs)

Authorization: Bearer eyJhbGciOiJSUzI1NiIs...

# Access token: 15 min, stateless, contains user ID + roles
# Refresh token: 7 days, stored server-side, one-time use (rotate on refresh)
# Token refresh endpoint:
POST /api/v1/auth/refresh
{ "refresh_token": "rt_abc123" }

OAuth2 (third-party integrations)

Use Authorization Code flow with PKCE for SPAs and mobile apps. Never use Implicit flow (deprecated). Client Credentials flow for service-to-service.

Rate Limiting

# Response headers (standard)
X-RateLimit-Limit: 100
X-RateLimit-Remaining: 42
X-RateLimit-Reset: 1708300800     # Unix timestamp when window resets
Retry-After: 30                    # seconds (on 429 response)

Implementation patterns:

• Fixed window: Simple but bursty at window boundaries. 100 req/min resets at :00.
• Sliding window: Smoother. Count requests in the last 60 seconds.
• Token bucket: Best for APIs. Allows bursts up to bucket size, refills at steady rate.

Tiered limits: Different limits per plan (free: 100/hr, pro: 10,000/hr). Different limits per endpoint (auth: 5/min, reads: 1000/min, writes: 100/min).

BAD/GOOD rate limiting patterns (the GOOD example is a fixed window counter; use a Lua script or SET key 1 EX 3600 NX so the count and expiry are atomic):

// BAD: Client-side only enforcement (easily bypassed)
if (requestCount > 100) {
  alert("Rate limit exceeded");
  return;
}

// GOOD: Server-side counter in Redis
const key = `rate:${userId}:${endpoint}`;
const count = await redis.incr(key);
if (count === 1) await redis.expire(key, 3600); // 1 hour window
if (count > 100) {
  res.set("Retry-After", "3600");
  return res.status(429).json({ error: "Rate limit exceeded" });
}

OpenAPI Spec (snippet)

openapi: 3.1.0
info:
  title: My API
  version: 1.0.0
paths:
  /api/v1/users:
    get:
      summary: List users
      operationId: listUsers
      tags: [users]
      parameters:
        - name: page
          in: query
          schema: { type: integer, minimum: 1, default: 1 }
        - name: per_page
          in: query
          schema: { type: integer, minimum: 1, maximum: 100, default: 25 }
      responses:
        "200":
          description: Paginated list of users
          content:
            application/json:
              schema:
                type: object
                properties:
                  data:
                    type: array
                    items: { $ref: "#/components/schemas/User" }
                  pagination:
                    $ref: "#/components/schemas/Pagination"
        "401":
          $ref: "#/components/responses/Unauthorized"
components:
  schemas:
    User:
      type: object
      required: [id, name, email]
      properties:
        id: { type: integer, format: int64 }
        name: { type: string, minLength: 1, maxLength: 100 }
        email: { type: string, format: email }
        created_at: { type: string, format: date-time }
  securitySchemes:
    bearerAuth:
      type: http
      scheme: bearer
      bearerFormat: JWT
security:
  - bearerAuth: []

Write spec first (design-first), then implement. Generate client SDKs and server stubs from the spec. Validate requests/responses against the spec in tests.

GraphQL: Schema Design

type Query {
  user(id: ID!): User
  users(first: Int = 20, after: String, filter: UserFilter): UserConnection!
}

type Mutation {
  createUser(input: CreateUserInput!): CreateUserPayload!
  updateUser(id: ID!, input: UpdateUserInput!): UpdateUserPayload!
}

type User {
  id: ID!
  name: String!
  email: String!
  orders(first: Int = 10, after: String): OrderConnection!
}

# Relay-style pagination
type UserConnection {
  edges: [UserEdge!]!
  pageInfo: PageInfo!
  totalCount: Int!
}

type UserEdge {
  cursor: String!
  node: User!
}

type PageInfo {
  hasNextPage: Boolean!
  hasPreviousPage: Boolean!
  startCursor: String
  endCursor: String
}

# Input types for mutations
input CreateUserInput {
  name: String!
  email: String!
}

# Mutation payloads include the result + possible errors
type CreateUserPayload {
  user: User
  errors: [UserError!]!
}

type UserError {
  field: String!
  message: String!
}

GraphQL design rules:

• Use Connection pattern (Relay spec) for all lists.
• Mutations take a single input argument and return a Payload type.
• Never expose database IDs directly; use opaque global IDs.
• Make fields non-nullable by default (String!). Only use nullable for fields that can genuinely be absent.

GraphQL: N+1 Problem

// BAD: each user resolver fetches orders individually
// 1 query for users + N queries for each user's orders = N+1

// GOOD: DataLoader batches and deduplicates
import DataLoader from "dataloader";

const ordersByUserLoader = new DataLoader(async (userIds) => {
  // Single query: SELECT * FROM orders WHERE user_id IN (...)
  const orders = await db.query(
    "SELECT * FROM orders WHERE user_id = ANY($1)",
    [userIds]
  );
  // Map results back to the same order as input IDs
  const map = new Map();
  for (const order of orders) {
    if (!map.has(order.user_id)) map.set(order.user_id, []);
    map.get(order.user_id).push(order);
  }
  return userIds.map((id) => map.get(id) || []);
});

// Resolver
const resolvers = {
  User: {
    orders: (user, _args, ctx) => ctx.loaders.ordersByUser.load(user.id),
  },
};

Always use DataLoader (or equivalent) for any field that fetches related data. Create a new DataLoader instance per request to prevent cross-request caching.

GraphQL: Subscriptions

type Subscription {
  orderStatusChanged(orderId: ID!): Order!
  newMessage(channelId: ID!): Message!
}

// Server (graphql-ws protocol, NOT the deprecated subscriptions-transport-ws)
import { WebSocketServer } from "ws";
import { useServer } from "graphql-ws/use/ws"; // graphql-ws v5 and older: "graphql-ws/lib/use/ws"

const wsServer = new WebSocketServer({ server: httpServer, path: "/graphql" });
useServer(
  {
    schema,
    context: async (ctx) => {
      // Authenticate on connection_init
      const token = ctx.connectionParams?.token;
      const user = await verifyToken(token);
      if (!user) throw new Error("Unauthorized");
      return { user };
    },
  },
  wsServer
);

Subscription tips: Use graphql-ws library (not the legacy subscriptions-transport-ws). Authenticate on WebSocket connection_init, not per message. Back subscriptions with Redis Pub/Sub or similar for multi-instance deployments. Always set connection timeouts and heartbeats.