llm-application-engineering

Engineering practices for building production LLM applications — model selection, structured outputs, streaming, prompt caching, retry logic, cost and latency control, observability, and the boring infrastructure that determines whether an LLM feature ships or stays a demo. Use this skill whenever the task involves calling an LLM API from application code, designing the architecture of an LLM-powered feature, debugging flaky or expensive LLM calls, hardening a prototype for production, choosing between models or providers, or any work where someone says "we have a prompt that works, now we need to ship it." Trigger even when the user does not explicitly say "production" — most LLM code becomes production code, and the practices here apply from day one.