AIO APEX
Claude Sonnet 4.6 or GPT-5A backend developer is wiring up a new payment provider integration before a Monday demo. She has the API docs open but needs to make sure she is not missing the error handling edge cases or webhook verification details that will cause production incidents later.Developer Tools

La Especificación de Integración de API: Convierte Cualquier API de Terceros en un Plan de Implementación Listo para Producción

Compartir:
La Especificación de Integración de API: Convierte Cualquier API de Terceros en un Plan de Implementación Listo para Producción

Why this prompt matters

Most API integration failures are not caused by calling the wrong endpoint — they happen because of unhandled error codes, missing retry logic, incorrect webhook signature verification, or silent rate-limit failures discovered only in production. A post-incident analysis of these bugs costs 10-50x more than preventing them. This prompt surfaces the integration gotchas upfront — the knowledge that normally requires either a painful incident or a colleague who has already burned their hand on this specific API.

What we use it for

A backend developer is wiring up a new payment provider integration before a Monday demo. She has the API docs open but needs to make sure she is not missing the error handling edge cases or webhook verification details that will cause production incidents later.

Prompt

Act as a senior software engineer with extensive experience integrating third-party APIs into production systems. You specialize in identifying edge cases, failure modes, and the "gotchas" that developers only discover after a painful production incident.

API to integrate: [API NAME, e.g., Stripe Payments, Twilio SMS, Salesforce, SendGrid]
My application: [DESCRIBE YOUR APP IN 1-2 SENTENCES]
What I need to do: [DESCRIBE THE SPECIFIC FUNCTIONALITY, e.g., "Process one-time payments and store customer payment methods for recurring billing"]
Tech stack: [LANGUAGE/FRAMEWORK, e.g., Node.js + Express, Python + Django, Go + Gin]
Known constraints: [ANY KNOWN LIMITS, e.g., "We are on the free tier", "Must be GDPR compliant", or "none"]

Generate a complete API integration spec with the following sections:

1. AUTHENTICATION SETUP
   - How authentication works for this API (API keys, OAuth 2.0, JWT, mTLS, etc.)
   - Where and how to store credentials securely in [TECH STACK]
   - Token refresh logic if required, including expiry handling

2. CORE IMPLEMENTATION PLAN
   - The exact sequence of API calls needed to fulfill [WHAT I NEED TO DO]
   - Required vs. optional parameters for each call
   - The shape of the request payload and what a successful response looks like
   - Any stateful dependencies between calls (e.g., must create customer before charging)

3. ERROR HANDLING MAP
   - The most common error codes this API returns and their root causes
   - Which errors are safe to retry vs. which are fatal and require user action
   - Recommended retry strategy: backoff timing, max attempts, jitter
   - How to surface errors to the user vs. handle silently

4. RATE LIMITS & QUOTAS
   - Request limits per second, minute, and day
   - How to detect when you are approaching limits (headers, error codes)
   - Recommended approach for high-volume usage (queuing, batching, caching)

5. WEBHOOKS & ASYNC EVENTS (if applicable)
   - What events this API delivers via webhook and their payload structure
   - How to verify webhook signatures to prevent spoofing
   - Idempotency: how to handle duplicate webhook deliveries safely

6. PRE-LAUNCH TESTING CHECKLIST
   - How to test this integration in sandbox or test mode without real data
   - The 5 most critical edge cases to validate before going to production
   - What metrics to monitor after launch (latency p95, error rate, success rate)

Constraints:
- Be specific to [API NAME]. Use real endpoint names, real error codes, and real response field names where known.
- Explicitly flag anything that commonly trips up developers integrating this API for the first time.
- If there are known quirks, undocumented behaviors, or common pitfalls with this API, mention them explicitly.

Result

API Integration Spec: Stripe Payments API
Application: SaaS subscription platform (Node.js + Express)
Goal: Process one-time payments and store payment methods for monthly recurring billing

1. Authentication Setup

  • Stripe uses separate secret and publishable API keys. Secret key (sk_live_...) stays server-side only — never exposed to the browser. Publishable key (pk_live_...) is safe for the frontend.
  • Store STRIPE_SECRET_KEY and STRIPE_WEBHOOK_SECRET in environment variables, never in source code. In Node.js: process.env.STRIPE_SECRET_KEY.
  • No token refresh needed — Stripe API keys do not expire. Rotate them via the Stripe Dashboard if compromised.

2. Core Implementation Plan

  • Step 1 — Create a Customer: POST /v1/customers with email and metadata. Store the returned customer.id in your database. This must happen before charging.
  • Step 2 — Collect payment method securely: Use Stripe.js on the frontend to tokenize card details. Your server never sees raw card numbers. The frontend returns a paymentMethod.id.
  • Step 3 — Create a PaymentIntent: POST /v1/payment_intents with amount (in cents), currency, customer, and payment_method. Set confirm: true to charge immediately.
  • Step 4 — Attach payment method for recurring billing: After first successful payment, call POST /v1/payment_methods/:id/attach with customer ID. Set as default: POST /v1/customers/:id with invoice_settings.default_payment_method.

3. Error Handling Map

  • card_declined — Card issuer rejected. Fatal; show user a human-readable message. Do NOT retry.
  • insufficient_funds — Fatal for this attempt. Prompt user to use a different card.
  • rate_limit (HTTP 429) — Safe to retry with exponential backoff: 1s, 2s, 4s, max 3 attempts.
  • idempotency_key_in_use — You sent the same idempotency key for a different request body. Use a new UUID.
  • Network timeout — Always use idempotency keys on POST requests so retries are safe. Check PaymentIntent status before retrying to avoid double charges.

4. Rate Limits

  • Default: 100 read requests/second, 100 write requests/second per account.
  • Approaching limits: Stripe returns X-RateLimit-Remaining and X-RateLimit-Reset headers. Watch for HTTP 429.
  • For high volume: use idempotency keys and consider Stripe's batch API for bulk operations.

5. Webhooks

  • Critical events to handle: payment_intent.succeeded, payment_intent.payment_failed, invoice.payment_failed (for subscriptions), customer.subscription.deleted.
  • Always verify the webhook signature: stripe.webhooks.constructEvent(payload, sig, process.env.STRIPE_WEBHOOK_SECRET). Reject any webhook that fails verification — forged webhooks are a real attack vector.
  • Make all webhook handlers idempotent. Stripe may deliver the same event twice. Use the event id to deduplicate.

6. Pre-Launch Testing Checklist

  • Use test card 4242 4242 4242 4242 for successful charges; 4000 0000 0000 9995 to simulate insufficient funds.
  • Test cases: successful charge → subscription creation → card decline → webhook delivery → duplicate webhook (send same event twice via Stripe Dashboard).
  • Monitor in production: payment success rate (alert if drops below 95%), PaymentIntent creation latency p95, webhook processing latency, and payment_intent.payment_failed rate by failure reason.

First-time Stripe gotcha: If you confirm a PaymentIntent server-side and the charge requires 3D Secure authentication, you will get a requires_action status instead of succeeded. Your frontend must handle this with stripe.handleNextAction(). Most developers hit this only when testing with European test cards.

Integrar una API de terceros por primera vez rara vez es el problema. El problema es todo lo que ocurre después: la respuesta de límite de tasa que olvidaste manejar, la verificación de firma de webhook que omitiste, el caso límite donde un tiempo de espera de red te deja sin saber si el pago se procesó. Los errores que causan incidentes de producción casi nunca son «llamé al endpoint equivocado» — son «no conocía este código de error específico» o «asumí que el webhook llegaría exactamente una vez».

Este prompt genera una especificación de integración completa antes de que escribas una línea de código. Revela los códigos de error, la lógica de reintento y las trampas de webhook que normalmente solo aprenderías de la experiencia o de una revisión posterior a un incidente.

Qué Produce el Prompt

El resultado es una especificación de integración estructurada en seis secciones que cubre la configuración de autenticación, la secuencia exacta de llamadas que requiere tu caso de uso, un mapa de manejo de errores por código de error, detalles de límites de tasa, verificación de firma de webhook y una lista de verificación previa al lanzamiento de los cinco casos límite con mayor probabilidad de causar incidentes de producción.

La restricción clave incorporada en el prompt es la especificidad: exige explícitamente nombres de endpoints reales, códigos de error reales y nombres de campos de respuesta reales — no consejos genéricos de integración de API. Para APIs ampliamente utilizadas como Stripe, Twilio, Salesforce, SendGrid, Slack y GitHub, los modelos actuales tienen suficientes datos de entrenamiento para producir resultados genuinamente útiles y específicos de la API. Para APIs menos comunes, pega una descripción del mecanismo de autenticación y un par de endpoints clave, y el modelo completará el resto a partir del contexto.

Cuándo Usarlo

Ejecuta este prompt antes de comenzar cualquier nueva integración de API — no después de que ya hayas escrito el camino feliz. El objetivo es aprender los requisitos de manejo de errores antes de que se conviertan en sesiones de depuración. Es particularmente valioso para APIs con flujos de autenticación complejos (OAuth 2.0 con refresh tokens, mTLS), APIs que entregan resultados asíncronos a través de webhook y APIs con requisitos de idempotencia no obvios.

Cómo Personalizarlo

Los cinco campos entre corchetes en la parte superior impulsan todo. Cuanto más específico sea tu campo «lo que necesito hacer», más dirigida será la secuencia de llamadas. Si estás integrando una API menos común, agrega una nota en «Restricciones conocidas» con cualquier nombre de endpoint o detalle de autenticación que ya conozcas — mejora significativamente la especificidad del resultado.

Para equipos que realizan trabajos de integración regulares, guarda el resultado como un archivo de especificación Markdown junto con el código de integración. Se convierte en documentación útil y una lista de verificación para la revisión de código.

developer toolsai-promptscodingbackendapi-integration
Compartir: