strands-agents/harness-sdk ? reverse-engineered prompt

Build me an open source SDK for building AI agents in Python and TypeScript, with a simple harness that gives me a fully working agent out of the box. I want it to run in my own app, not depend on a hosted control plane, and work with different model providers so I can switch between services without rewriting everything.

Please include the basics a real agent needs, like tools, structured output, memory, sessions, streaming, guardrails, tracing, and ways to stop or limit the agent so it does not run forever. It should also support MCP and multi agent setups. Make the Python and TypeScript versions feel consistent, and add a small CLI so I can chat with an agent from the terminal.

If it helps, look up current docs online for any provider or SDK details. I’d also like a clean docs site and a few good examples so someone can get started quickly.