bsamud/openfoundry-agentic-framework — reverse-engineered prompt

Build me a Python framework called OpenFoundry for running teams of AI agents from one place. I want someone to be able to install it, start a server, use a simple command line tool, and send work to agents through an HTTP API.

The main idea is that users can define workflows where different agents handle planning, coding, quality checks, deployment, and release tasks. The workflow runner should understand which steps depend on each other and run independent steps in parallel when possible. It should support more than one LLM provider, pick models based on simple rules, and fall back to another provider if one fails.

Please include safety checks for things like prompt injection and personal information, plus basic monitoring with traces and metrics. Add a clean project structure, config files, Docker support, readable docs, and a quick start showing install, initialize, serve, and list agents. Keep it practical and usable, not just a demo.