agents-openailisted

# Agents on OpenAI Platform Patterns OpenAI's agent-building surface centers on a few primitives: chat models with native function calling, structured outputs, persistent threads for conversation state, and file/code-interpreter tools. This skill covers using them as an agent substrate. ## Overview The core loop: the model reasons, emits function calls with structured arguments, your code executes them, results go back into the conversation, repeat. Structured outputs (JSON schema mode) make the model's responses machine-readable. Threads persist conversation state server-side so you don't manage history yourself. Built-in tools — code execution, file search — cover common needs without custom tooling. Your architecture decisions sit on top: when to call functions vs. answer, how to manage the loop, where humans intervene. ## When to use - Building an agent or assistant powered by OpenAI models. - Needing reliable function calling with validated arguments. - Prototyping quickly with managed threads and built-in tools before custom infrastructure. - Migrating from raw chat completions to a structured agent loop. ## Core concepts - **Function calling**: the model emits structured calls to your functions; you execute and return results. Define functions with clear names, descriptions, and JSON schemas. - **Structured outputs**: schema-constrained generation — the model must return valid JSON matching your schema. Eliminates parsing hacks for data extraction and t