ai-integration-pro
AI Integration (professional)
Skill text is English; answer in the user’s preferred language when Cursor User Rules or the conversation specify it (code and API names stay as in docs).
Use official Anthropic docs and OpenAI docs for API truth; align production expectations with conversation state, evaluation, safety layers, and observability — not only prompts. This skill encodes prompt engineering discipline, tool orchestration, RAG architecture, and operational reliability. Confirm provider/model, SDK version, context limits, and streaming vs batch when known.
Boundary
ai-integration-pro owns LLM integration architecture: prompts, tools, streaming, RAG query+ingest design, memory/trim policy, routing/fallback, eval hooks, observability contracts, and minimum safety integration. It does not replace security-pro for full abuse programs, deployment-pro for infra, or vendor-specific compliance sign-off.
| Skill | When to combine |
|---|---|
security-pro |
API keys, injection, output policy, data classification, moderation programs |
caching-pro |
Prompt caching, embedding caches, memoization |
api-design-pro |
AI-powered HTTP/WebSocket API design |
testing-pro |
Mock LLMs, golden sets, CI eval |
postgresql-pro |
pgvector, RLS on embedding rows |