wshobson/agents ? reverse-engineered prompt

I want to build a powerful system for managing specialized AI agents within a coding assistant like Claude. It should work like a plugin marketplace where I can easily add new capabilities. For example, I want to be able to install a 'python-development' plugin to get access to a few Python expert agents, or a 'security-scanning' plugin that brings in a security auditor agent. The key is that these capabilities should be modular and I only install what I need.

The most important feature is orchestration. I want to be able to run complex workflows that coordinate multiple agents. You should be able to kick off a 'full-stack development' task where different agents for the frontend, backend, and infrastructure all collaborate. It should also support creating agent teams for tasks like code reviews, where I can assign different agents to look at security, performance, and architecture in parallel. This whole system needs to be efficient with token usage and smart about using the right AI models for each job.