Agent Skills: Api Development Expert

Api Development Expert

When designing REST APIs, follow these core architectural principles:

Resource-Oriented Design

• Use nouns for resources (plural form): /users, /products, /orders
• Avoid verbs in URIs: ❌ /getUsers, /createProduct
• Structure hierarchically: /users/{userId}/orders (orders belonging to a user)
• Use lowercase with hyphens: /product-details not /productdetails
• No trailing slashes: /users not /users/

HTTP Methods (Verbs with Purpose)

• GET - Retrieve resources (idempotent & safe, no side effects)
• POST - Create new resources (not idempotent, returns 201 Created with Location header)
• PUT - Replace entire resource or upsert (idempotent)
• PATCH - Partial update (not idempotent, use application/json-patch+json)
• DELETE - Remove resource (idempotent, returns 204 No Content or 200 OK)

Query Parameters for Filtering, Sorting, and Pagination

• Filtering: /products?category=electronics&price_gt=100
• Sorting: /products?sort_by=price&order=desc
• Pagination: /products?page=2&limit=10
  • Use offset-based (simple but inefficient for deep pages) or cursor-based (efficient for large datasets)

API Versioning Strategies

Choose one and stick to it:

• URI Versioning (Most common): /v1/users, /api/v2/products
  • Simple for clients, but makes URIs less clean
• Header Versioning: Accept: application/vnd.myapi.v1+json
  • Cleaner URIs, but slightly complex for caching and some clients
• Content Negotiation: Use Accept header to specify desired media type and version

OpenAPI/Swagger Specification

Use OpenAPI 3.0+ to define your API specification:

Benefits:

• Machine-readable API specification
• Auto-generates interactive documentation portals
• Client SDK generation
• Request/response schema validation
• IDE and API tool auto-validation

Define schemas for:

• Request parameters (required fields, allowed values, data types)
• Response structures
• Error responses
• Authentication methods
• Enum lists for restricted values

Example: Define validation rules so invalid requests are caught before reaching your backend

Rate Limiting Patterns

Protect against abuse and ensure fair usage:

Implementation strategies:

• Use 429 Too Many Requests status code
• Return rate limit headers:
  • X-RateLimit-Limit: 1000
  • X-RateLimit-Remaining: 999
  • X-RateLimit-Reset: 1640000000
• Common patterns:
  • Fixed window: Simple but allows bursts at boundaries
  • Sliding window: More accurate, prevents boundary gaming
  • Token bucket: Allows controlled bursts
  • Leaky bucket: Smooths out traffic

Error Handling Standards

Consistent Error Response Structure:

{
  "error": {
    "code": "validation_error",
    "message": "Input validation failed.",
    "details": [{ "field": "email", "message": "Invalid email format." }]
  }
}

Use Appropriate HTTP Status Codes:

2xx Success: 200 OK, 201 Created, 204 No Content

3xx Redirection: 301 Moved Permanently, 304 Not Modified

4xx Client Error:

• 400 Bad Request - General client error
• 401 Unauthorized - Authentication missing/failed
• 403 Forbidden - Authenticated but no permission
• 404 Not Found - Resource doesn't exist
• 405 Method Not Allowed - Invalid HTTP method
• 409 Conflict - Resource already exists
• 422 Unprocessable Entity - Semantic validation error
• 429 Too Many Requests - Rate limiting

5xx Server Error:

• 500 Internal Server Error - Generic server error
• 503 Service Unavailable - Service temporarily down

Provide machine-readable codes AND human-readable messages

Authentication Patterns

OAuth 2.1 (Industry standard for delegated authorization)

• Mandatory PKCE for all authorization code flows
• Authorization Code + PKCE flow for SPAs, mobile, and web apps
• Removed flows: Implicit grant and Resource Owner Password Credentials (security risks)
• Exact redirect URI matching (no wildcards)
• Never send bearer tokens in query strings (use Authorization header)
• Implement refresh token rotation or sender-constrained tokens

JWT (JSON Web Tokens) for stateless authentication:

• Short expiry times (≤15 minutes for access tokens)
• Use refresh tokens for long-lived sessions
• Include claims for authorization decisions
• Validate signature, expiry, and issuer

API Keys for simpler integrations:

• Use for service-to-service authentication
• Rotate regularly
• Never expose in client-side code
• Implement rate limiting per key

Performance & Caching

HTTP Caching Headers:

• Cache-Control: max-age=3600 - Cache for 1 hour
• ETag - Entity tag for conditional requests
• Expires - Absolute expiration time
• 304 Not Modified - Return for unchanged resources

Caching strategies:

• Client-side caching (browser cache)
• Proxy/CDN caching (intermediate caches)
• Server-side caching (database query cache, object cache)

Optimization techniques:

• Compression: Use GZIP for large responses
• Pagination: Return only needed data
• Field selection: Allow clients to request specific fields (?fields=id,name)
• Async operations: For long-running tasks, return 202 Accepted with status endpoint

API Documentation Best Practices

Comprehensive documentation must include:

• Overview and getting started guide
• Authentication and authorization details
• Endpoint descriptions with HTTP methods
• Request parameters and body schemas
• Response structures with examples
• Error codes and messages
• Rate limits and usage policies
• SDKs and client libraries
• Changelog for version updates

Use tools:

• Swagger UI / OpenAPI for interactive docs
• Postman collections for testing
• Code examples in multiple languages

OpenAPI 3.1 (2026 Best Practice)

Upgrade from OpenAPI 3.0 to 3.1 for full JSON Schema 2020-12 compliance:

Key differences from 3.0:

| Feature | OpenAPI 3.0 | OpenAPI 3.1 | | ----------------------- | --------------------- | -------------------------- | | JSON Schema compliance | Subset + extensions | Full JSON Schema 2020-12 | | Nullable fields | nullable: true | type: ["string", "null"] | | Webhooks | Not supported | webhooks object at root | | $schema declaration | Not allowed | Allowed at document root | | example vs examples | Both allowed together | Mutually exclusive |

OpenAPI 3.1 Webhook definition:

openapi: '3.1.0'
info:
  title: My API
  version: '1.0.0'

webhooks:
  orderCreated:
    post:
      requestBody:
        required: true
        content:
          application/json:
            schema:
              $ref: '#/components/schemas/OrderEvent'
      responses:
        '200':
          description: Webhook received successfully

OpenAPI Overlays (v1.1.0, Jan 2026):

Overlays are a companion spec that applies targeted transformations to an OpenAPI document without modifying the source. Use cases:

• Adding environment-specific server URLs
• Injecting partner-specific authentication metadata
• Removing internal endpoints before publishing externally

# overlay.yaml
overlay: '1.0.0'
info:
  title: Partner API Overlay
  version: '1.0.0'
actions:
  - target: "$.paths['/internal/**']"
    remove: true
  - target: '$.info'
    update:
      x-partner-id: 'acme-corp'

OpenAPI 3.2 (September 2025) adds:

• Hierarchical tag metadata (summary, parent, kind)
• Streaming-friendly media types with itemSchema
• query HTTP operations and querystring parameters
• OAuth2 device flow and metadata URL support

API Versioning Strategies (2026)

Choose and document your versioning strategy before the first public release:

URI versioning (most common, most explicit):

GET /v1/users
GET /v2/users

• Easy to test, cache, and route
• Consider: version only on breaking changes (additive changes are backward compatible)

Header versioning:

Accept: application/vnd.myapi.v2+json

• Cleaner URIs; slightly harder for browser-based testing

Deprecation lifecycle:

Deprecation: true
Sunset: Sat, 01 Jan 2027 00:00:00 GMT
Link: <https://api.example.com/v2/users>; rel="successor-version"

GraphQL schema evolution (no URI versioning needed):

• Use @deprecated(reason: "Use newField instead") to phase out fields
• Add new fields without removing old ones (additive-only changes)
• Never rename or remove fields from a live schema; deprecate and provide migration path

Rate Limiting — Advanced Patterns (2026)

Standardized headers (IETF draft draft-ietf-httpapi-ratelimit-headers):

RateLimit-Limit: 100
RateLimit-Remaining: 87
RateLimit-Reset: 1
Retry-After: 30

Algorithm comparison:

| Algorithm | Burst Handling | Accuracy | Use Case | | ---------------------- | ---------------------- | -------- | ---------------- | | Fixed window | Allows boundary bursts | Low | Simple rate caps | | Sliding window log | No burst | High | Strict SLAs | | Sliding window counter | Small burst | Medium | General purpose | | Token bucket | Controlled burst | High | API gateways | | Leaky bucket | No burst (smoothed) | High | Traffic shaping |

Tiered rate limiting — differentiate by caller type:

rate_limits:
  anonymous: 60/hour
  authenticated: 1000/hour
  premium: 10000/hour
  internal_service: unlimited

HATEOAS & Richardson Maturity Model

Design APIs for discoverability — include hypermedia links so clients can navigate state transitions:

Richardson Maturity Levels:

| Level | Description | Example | | ----- | ------------------------------- | --------------------------------- | | 0 | Single endpoint (RPC over HTTP) | POST /api with action field | | 1 | Resource-based URIs | GET /users/123 | | 2 | HTTP verbs + status codes | DELETE /users/123 returns 204 | | 3 | HATEOAS (hypermedia controls) | Response includes _links |

Level 3 example response:

{
  "id": "order-42",
  "status": "pending",
  "total": 99.99,
  "_links": {
    "self": { "href": "/orders/42" },
    "confirm": { "href": "/orders/42/confirm", "method": "POST" },
    "cancel": { "href": "/orders/42/cancel", "method": "DELETE" },
    "customer": { "href": "/users/7" }
  }
}

Level 3 is aspirational; most production APIs operate at Level 2 and selectively add hypermedia links for complex workflows.

GraphQL Federation (2026)

For microservices architectures using GraphQL, federation enables a unified supergraph from distributed subgraphs:

Core concepts:

• Supergraph: The combined schema exposed to clients
• Subgraph: Individual service schemas that each own their type definitions
• Router: Federates queries across subgraphs (Apollo Router, WunderGraph Cosmo)

Best practices:

• One entity (e.g., User) can be defined in its owning subgraph and extended in others using @key and @extends
• Schema governance: use a schema registry (Apollo Studio, Cosmo Schema Registry) to validate changes before deployment
• AI/LLM traffic reshaping architecture requirements in 2026 — plan for high-volume, streaming-friendly subgraph operations

Iron Laws

1. ALWAYS version your API from day 1 — never introduce breaking changes without a version bump; use URI versioning (/v1/, /v2/) so clients can migrate on their schedule.
2. NEVER return 200 OK for errors — use proper HTTP status codes: 400 (bad request), 401 (unauthorized), 403 (forbidden), 404 (not found), 422 (validation failed), 500 (server error).
3. ALWAYS document every endpoint in OpenAPI 3.1 — undocumented APIs cannot be safely consumed; OpenAPI 3.1 provides JSON Schema 2020-12 compliance and webhook support.
4. NEVER include sensitive data in error responses — stack traces, database schema, and internal file paths are attack vectors; return only machine-readable error codes and safe messages.
5. ALWAYS implement rate limiting on all public endpoints — unauthenticated endpoints without rate limiting are DoS vectors; respond with 429 and Retry-After header.

Anti-Patterns

| Anti-Pattern | Why It Fails | Correct Approach | | ------------------------------------------ | --------------------------------------------------------------- | ------------------------------------------------ | | Verbs in URIs (/getUser, /createOrder) | Violates REST constraints; HTTP method conveys the verb | Use nouns: /users, /orders with GET/POST | | No API versioning from day 1 | Breaking changes instantly break all existing clients | URI versioning: /v1/resource from the start | | Returning 200 OK for errors | Clients can't distinguish success from failure programmatically | Use correct HTTP status codes | | No rate limiting on public endpoints | DoS vulnerability; single client can exhaust resources | Rate limit with X-RateLimit-* headers + 429 | | Leaking server internals in errors | Stack traces and DB errors are attack vectors | Return error codes + safe messages only | | No OpenAPI specification | Clients must guess request/response format | Document all endpoints in OpenAPI 3.1 |

Consolidated Skills

This expert skill consolidates 1 individual skills:

• api-development-expert

Memory Protocol (MANDATORY)

Before starting:

cat .claude/context/memory/learnings.md

After completing: Record any new patterns or exceptions discovered.

ASSUME INTERRUPTION: Your context may reset. If it's not in memory, it didn't happen.