Yaatal-labs/Yaatal-Harness ? reverse-engineered prompt

Build me the Yaatal Harness as a Rust workspace for the AI control plane of the system.

I want a working foundation where agent actions go through a single governed path, with audit records, policy checks, tool execution, memory access, and evaluation all wired together. The current repo already has the core crates and some runner and proposal flow, so please fill in the missing pieces and make the whole thing feel like a real working product, not just a scaffold.

Keep it CLI first for tools, with clear help output, JSON friendly commands, and safe defaults. Make sure actions are traced, policy is enforced before anything risky runs, and there is a simple loop that can collect metrics, score runs, and suggest configuration changes for human review. If anything needs current crate or Rust guidance, look it up online.

Also clean up the docs and examples so it is easy to understand how the Harness connects to the Engine without mixing the two runtimes together.