jpicklyk/task-orchestrator ? reverse-engineered prompt

Build me an MCP server in Kotlin that acts like a task orchestrator for AI agents.

I want it to keep persistent work items in a database, let me create tasks with child tasks and dependencies, and stop an agent from moving a task forward unless the required notes or checks are done. It should support actor attribution too, so every update can tell which agent made the change, and if auditing is turned on, it should reject writes from anonymous agents.

Please make it work with any MCP compatible client, and include a simple way to define workflow rules in YAML so different task types can require different notes or quality gates. I also want commands for creating work trees, updating items, querying notes, getting the current context, and searching across items and notes. If there are tests and Docker files already in the repo, wire everything up so it runs cleanly and is easy to start locally. If you need to look up current MCP docs online, do that.