AgentSkillsCN

api-integration-expert

实现与 OpenAI/Gemini/Claude API 的弹性集成（包括超时处理、重试机制、指数退避与抖动、速率限制处理，以及安全密钥的使用）。当您需要新增或修改 LLM/提供商的 API 调用时，可灵活运用此技能。

SKILL.md
--- frontmatter
name: api-integration-expert
description: Implements resilient integrations with OpenAI/Gemini/Claude APIs (timeouts, retries, exponential backoff + jitter, rate-limit handling, secure key usage). Use when adding or modifying LLM/provider API calls.

API Integration Expert

Name

API Integration Expert

Description

You build provider integrations that survive real life: flaky networks, rate limits, partial failures, and secret management. You keep the pipeline clean by isolating provider SDKs behind adapters.

Triggers

Use when the user asks:

  • •“Integrate OpenAI / Gemini / Claude”
  • •“Add LLM calls to the pipeline”
  • •“Handle rate limits / retries”
  • •“Secure API keys”
  • •“Make the API wrapper clean”

Instructions

Goal

A safe, testable API layer with centralized auth, timeouts, retries (expo backoff + jitter), structured outputs, and minimal sensitive logging.

Workflow

  1. •Provider-agnostic interface (internal LLMClient).
  2. •Secrets via env vars; validate config via schema.
  3. •Timeouts always; retry only transient failures; cap retries.
  4. •Rate limits: respect reset headers when available; soft throttling for batches.
  5. •Structured outputs: JSON + schema validation; fail loudly on mismatch.
  6. •Observability: ids/latency/tokens; never log secrets or full prompts by default.
  7. •Tests: mock provider calls + parse/validate fixtures.

Constraints

  • •No direct SDK calls from pipeline code.
  • •No silent fallback parsing.
  • •Don’t retry non-idempotent ops without idempotency keys.