openai-assistants

What it does

Guides a coding agent to work with OpenAI Assistants API v2 to build stateful chatbots using Assistants, Threads, Messages, and Runs, including tools like Code Interpreter, File Search, and Function Calling.

How it works

Describes object models (Assistants, Threads, Messages, Runs) and their interactions:

• Create an Assistant with name, instructions, model, tools, tool_resources, and metadata.
• Create/Retrieve/Update/Delete/List Assistants.
• Create Threads (empty or with initial messages) and retrieve/update/delete metadata.
• Add Messages to a Thread with role, content, attachments, and metadata.
• Create Runs on a Thread with assistant_id, optional instructions, additional_messages, and overrides for tools/temperature/top_p/max tokens; poll Run status until completion or requires_action; handle streaming and cancellation.
• Streaming Runs provide real-time events for run and message progress.
• Use Tools (Code Interpreter, File Search, Function Calling) within Runs and Messages as described.
• Use available outputs from Code Interpreter via listing messages and accessing file contents.

When to use it

Use when building stateful chatbots, implementing RAG, or troubleshooting issues related to thread activity, vector store delays, polling timeouts, or file upload errors.

What it can touch

Tools include Code Interpreter, File Search, and Function Calling as per the skill's guidance. The content shows how to reference tools in assistants and runs, and how to attach tools to messages and runs.

Caveats

OpenAI Assistants API v2 is sunset in H1 2026; migration to Responses API is advised. Deeper migration guidance is in the migration-to-responses path. Deactivation and deprecation status is noted, with recommended use only for existing code or legacy maintenance. The skill documents how to create, retrieve, update, delete, list, and run assistants, threads, messages, and runs, including polling patterns and streaming behavior.