aden-hive/hive ? reverse-engineered prompt

I want to build a platform for creating and running AI agents that can handle real-world business tasks, not just simple demos. The core idea is that I can describe a goal in plain English to a main "queen" AI, which then figures out the steps and creates a team of other agents to execute the plan.

These agents need to be robust. They should be able to run for a long time, and if one of them fails or crashes, the system should be able to recover and continue the job. It's also important that they can use a web browser to do things like find information or fill out forms. If an agent gets stuck, there should be a way for a human to step in and help. The whole system should be self-improving, learning from failures to adapt and get better over time.

Please build this with a web-based dashboard where I can manage everything. It should let me connect to different LLMs like OpenAI, Anthropic, and Google.