Api Design

API Design

Approach every API as a product. The developer calling your API is your user. An API that is hard to understand, inconsistent, or unpredictable is a broken product — even if it technically works.

Design the API before writing a single line of implementation. An API is a contract. Changing it after clients depend on it is expensive. Getting it right upfront is cheap.

---

API Design Principles

• Spec first, implement second. Write the OpenAPI spec, GraphQL schema, or proto file before any implementation. The spec is the contract. Implementation details are irrelevant to the caller.
• Consistency is the most important quality. Every endpoint in an API should behave according to the same rules: same error format, same naming convention, same pagination pattern, same auth mechanism. Inconsistency forces callers to write special-case code for every endpoint.
• APIs are forever. Every field you add is a field you must support until you version or sunset the API. Every field you remove is a breaking change. Design with permanence in mind.
• Design for the caller, not the data model. Your database schema is an implementation detail. Your API shape should reflect what callers need, not what your ORM produces.
• Explicit over implicit. Undefined behaviour in an API becomes the behaviour callers depend on. Make every behaviour explicit: required vs optional fields, error codes, rate limits, pagination behaviour at the end of a list.
• Developer experience is a first-class requirement. An API that requires a PhD to understand will not be adopted. Good error messages, consistent naming, and accurate documentation are as important as correct behaviour.

---

Step 0: Ground the API Design

Before designing any endpoint, answer these questions:

1. Who is the caller? Internal service, third-party developer, mobile app, browser, CLI? Each has different needs (auth mechanism, response size, error verbosity).
2. What resources does the API expose? List the core domain entities the API manages.
3. What operations does each resource support? CRUD is not always the right decomposition — sometimes cancel, approve, publish are better verbs than generic CRUD.
4. What are the read/write patterns? High-read, low-write? Real-time? Bulk operations? Streaming? This determines the API style.
5. What are the latency and payload size constraints? A mobile API on a 3G connection has different constraints than a backend-to-backend integration.
6. What versioning strategy is needed from day one? If this API will have external consumers, versioning must be designed in before the first endpoint ships.

---

Choosing the Right API Style

| Style | Best for | Avoid when | |-------|----------|------------| | REST | Resource-oriented APIs, public APIs, broad client compatibility, simple CRUD | Complex queries with many relationships, real-time, or highly variable response shapes | | GraphQL | Flexible queries, multiple clients with different data needs, frontend-driven development, deeply nested data | Simple APIs, teams without GraphQL tooling, when over-fetching is not a real problem | | gRPC | High-performance internal service communication, streaming, strongly-typed contracts, polyglot microservices | Browser clients (requires grpc-web proxy), teams unfamiliar with protobuf | | WebSocket | Real-time bidirectional communication (chat, live dashboards, multiplayer) | Request-response patterns that do not need real-time; adds complexity without benefit | | Webhooks | Asynchronous event notification to external systems | When the caller needs to poll or query state; use REST polling or SSE instead |

---

REST API Design Standards

URL Structure

# Pattern
/{version}/{resource}/{id}/{sub-resource}

# Examples — good
GET    /v1/users
GET    /v1/users/{userId}
GET    /v1/users/{userId}/orders
POST   /v1/users
PUT    /v1/users/{userId}
PATCH  /v1/users/{userId}
DELETE /v1/users/{userId}

# Actions that don't map to CRUD — use sub-resources
POST   /v1/orders/{orderId}/cancel
POST   /v1/invoices/{invoiceId}/send
POST   /v1/users/{userId}/password-reset

URL rules:

• Always lowercase, hyphen-separated (user-profiles, not userProfiles or user_profiles)
• Nouns for resources, not verbs (/users, not /getUsers)
• Plural for collections (/users, not /user)
• Version in the URL path (/v1/) for public APIs — query param or header for internal APIs
• Never expose database IDs directly where possible (use UUIDs or opaque string IDs)

HTTP Methods — Correct Usage

| Method | Semantics | Idempotent | Safe | |--------|-----------|------------|------| | GET | Retrieve resource(s) | ✅ | ✅ | | POST | Create a new resource or trigger an action | ❌ | ❌ | | PUT | Replace a resource entirely | ✅ | ❌ | | PATCH | Partially update a resource | ❌ (should be) | ❌ | | DELETE | Remove a resource | ✅ | ❌ |

Rule: GET requests must never have side effects. Never use GET to trigger a state change.

Request & Response Shape

Request body — always:

{
  "name": "Jane Doe",
  "email": "jane@example.com",
  "role": "admin"
}

Single resource response:

{
  "id": "usr_01HXYZ",
  "name": "Jane Doe",
  "email": "jane@example.com",
  "role": "admin",
  "createdAt": "2026-01-15T10:30:00Z",
  "updatedAt": "2026-01-15T10:30:00Z"
}

Collection response — always wrap in an envelope:

{
  "data": [
    { "id": "usr_01HXYZ", "name": "Jane Doe" },
    { "id": "usr_02HABC", "name": "John Smith" }
  ],
  "pagination": {
    "cursor": "eyJpZCI6InVzcl8wMkhBQkMifQ==",
    "hasMore": true,
    "total": 247
  }
}

Why an envelope? Adding metadata (pagination, request ID, warnings) to a bare array response is a breaking change. An envelope allows non-breaking additions forever.

Naming Conventions

• Field names: camelCase for JSON APIs (firstName, not first_name or FirstName)
• Timestamps: ISO 8601 UTC ("2026-01-15T10:30:00Z") — never Unix timestamps in the response body
• Booleans: positive framing (isActive, not isNotActive; isEnabled, not isDisabled)
• IDs: string type always, even if internally numeric (prevents JavaScript integer overflow)
• Money: integer cents, never floating-point ("amount": 1999 means $19.99)
• Enums: SCREAMINGSNAKECASE ("status": "IN_PROGRESS")

---

Error Response Standard

Every API must have one error format used consistently across all endpoints:

{
  "error": {
    "code": "VALIDATION_ERROR",
    "message": "Request validation failed",
    "details": [
      {
        "field": "email",
        "code": "INVALID_FORMAT",
        "message": "Must be a valid email address"
      }
    ],
    "requestId": "req_01HXYZ123",
    "docsUrl": "https://docs.example.com/errors/VALIDATION_ERROR"
  }
}

Rules:

• code is a machine-readable string constant — callers switch on this, not on the HTTP status
• message is human-readable — never put a machine-parseable value here
• details is an array — multiple validation errors in one response, never force callers to fix one error at a time
• requestId on every error response — this is how support traces the request in logs
• docsUrl for each error code — link to documentation explaining the error and how to fix it

HTTP Status Codes — Correct Usage

| Code | When to use | |------|------------| | 200 OK | Successful GET, PATCH, PUT | | 201 Created | Successful POST that creates a resource | | 204 No Content | Successful DELETE or action with no response body | | 400 Bad Request | Validation error, malformed request | | 401 Unauthorized | Not authenticated | | 403 Forbidden | Authenticated but not authorized for this resource | | 404 Not Found | Resource does not exist | | 409 Conflict | State conflict (duplicate, version mismatch) | | 422 Unprocessable Entity | Semantically invalid request (valid syntax, invalid business logic) | | 429 Too Many Requests | Rate limit exceeded | | 500 Internal Server Error | Unexpected server error | | 503 Service Unavailable | Planned downtime or dependency unavailable |

Never use 200 with an error body. { "success": false, "error": "..." } with a 200 status is a broken API.

---

Pagination Patterns

Cursor-based Pagination (recommended for most cases)

GET /v1/users?cursor=eyJpZCI6InVzcl8wMkhBQkMifQ==&limit=20

When: Ordered, append-heavy collections (feeds, logs, events). Stable pages even when new items are inserted.

Offset-based Pagination (simple, but has edge cases)

GET /v1/products?page=3&pageSize=20

When: Admin UIs where users jump to specific pages. Avoid for large or frequently-updated datasets (items shift as pages load).

Keyset Pagination (for high-performance sorted queries)

GET /v1/orders?afterId=ord_01HXYZ&limit=50

When: Database queries on an indexed column where offset queries become slow.

Pagination response fields (always include):

• cursor or nextPage — how to get the next page
• hasMore (boolean) — whether more results exist after this page
• total (optional) — total count (expensive on large datasets; omit if not needed)
• limit — the limit that was applied (echo it back)

---

API Versioning Strategy

URL Path Versioning (recommended for public APIs)

/v1/users
/v2/users

Pros: Explicit, easy to route in proxies/gateways, cacheable. Use for: External/public APIs, mobile app APIs (clients pin to a version).

Header Versioning (recommended for internal APIs)

API-Version: 2026-01-15

Pros: Keeps URLs clean; date-based versions are self-documenting. Use for: Internal services, APIs with sophisticated clients.

Breaking vs Non-Breaking Changes

Non-breaking (safe to add without versioning):

• Adding new optional fields to responses
• Adding new optional request parameters
• Adding new endpoints
• Loosening validation on existing fields

Breaking (requires version bump):

• Removing or renaming fields
• Changing field types
• Changing HTTP status codes
• Tightening validation
• Changing authentication requirements
• Removing endpoints

Deprecation policy: Mark deprecated fields with a X-Deprecated-Fields response header and a deprecated note in the OpenAPI spec. Give callers a minimum of 6 months notice before removal.

---

OpenAPI Spec Standards

Every REST API ships with an OpenAPI 3.x spec. No exceptions.

openapi: 3.1.0
info:
  title: User Management API
  version: 1.0.0
  description: Manages user accounts and profiles

paths:
  /v1/users/{userId}:
    get:
      summary: Get a user by ID
      operationId: getUserById
      tags: [Users]
      parameters:
        - name: userId
          in: path
          required: true
          schema:
            type: string
      responses:
        '200':
          description: User found
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/User'
        '404':
          $ref: '#/components/responses/NotFound'
        '401':
          $ref: '#/components/responses/Unauthorized'

components:
  schemas:
    User:
      type: object
      required: [id, name, email, createdAt]
      properties:
        id:
          type: string
          example: usr_01HXYZ
        name:
          type: string
          example: Jane Doe
        email:
          type: string
          format: email
        createdAt:
          type: string
          format: date-time

OpenAPI rules:

• operationId on every endpoint — used for SDK generation
• tags on every endpoint — groups endpoints in documentation
• Every response code documented — not just 200
• $ref for shared schemas — no duplication
• Realistic example values — not string, foo, or 123
• Security schemes defined and applied to every protected endpoint

---

Rate Limiting

Every public API must implement rate limiting. Communicate it clearly:

Response headers (always include on rate-limited APIs):

X-RateLimit-Limit: 1000
X-RateLimit-Remaining: 842
X-RateLimit-Reset: 1737892800
Retry-After: 60        (only on 429 responses)

Rate limit strategy:

• Per API key / per user: prevents one caller from starving others
• Per endpoint: expensive endpoints get tighter limits than cheap ones
• Sliding window preferred over fixed window (avoids burst at window boundary)
• Return 429 with Retry-After header — never silently drop requests

---

Authentication Standards

| Mechanism | Use case | |-----------|----------| | Bearer JWT | User-facing APIs, mobile/web clients | | API Key (Authorization: Bearer sk_...) | Server-to-server, third-party integrations | | OAuth 2.0 + PKCE | Third-party access on behalf of a user | | mTLS | High-security internal service communication |

Rules:

• Always use Authorization: Bearer header — never query parameters for auth tokens
• API keys in query parameters appear in server logs. Never.
• Short-lived access tokens (15 min – 1 hour) with refresh token rotation
• Scopes on API keys — a key should only have access to what it needs

---

Bundled Reference

Read [openapi-review.md](./references/openapi-review.md) when reviewing or writing an OpenAPI contract. It supplements, but does not replace, the API-specific requirements in this skill.

API Design Review Checklist

Before finalising any API design:

Resource & URL Design

• [ ] URLs are nouns, lowercase, hyphenated
• [ ] HTTP methods used correctly (no state-changing GETs)
• [ ] Version included in URL or header strategy defined
• [ ] IDs are string type (not integer)
• [ ] Actions not mappable to CRUD use sub-resource + POST

Request & Response

• [ ] Collections wrapped in an envelope object
• [ ] Consistent field naming (camelCase)
• [ ] Timestamps in ISO 8601 UTC format
• [ ] Money in integer cents, not floats
• [ ] All required fields documented

Errors

• [ ] Single error format used across all endpoints
• [ ] Machine-readable error codes defined
• [ ] Field-level validation errors returned in one response (not one at a time)
• [ ] Correct HTTP status codes used
• [ ] RequestId on every error response

Pagination

• [ ] All collection endpoints paginated (never return unbounded lists)
• [ ] Pagination strategy documented
• [ ] hasMore and cursor/nextPage in every paginated response

Versioning & Breaking Changes

• [ ] Versioning strategy defined before first endpoint ships
• [ ] Deprecation policy defined
• [ ] No breaking changes to existing endpoints without a version bump

Security

• [ ] Auth mechanism defined for all endpoints
• [ ] Rate limiting applied and communicated via headers
• [ ] No secrets in URLs or query parameters
• [ ] HTTPS only — no HTTP fallback

Documentation

• [ ] OpenAPI 3.x spec complete with all endpoints, schemas, and error responses
• [ ] Realistic examples on all request/response fields
• [ ] Authentication documented with example token format
• [ ] Rate limits documented

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: CODE-SAURABH
• Source: CODE-SAURABH/OpenSkills
• License: MIT

Install and usage instructions live in the source repository linked above.