LFX-Talent-Angels/TA-agents ? reverse-engineered prompt

Build me the Talent Angels assistant runtime for reasoning over skill, task, and occupation taxonomies with Graph RAG.

I want one main assistant that understands a user’s goal and can route to four capabilities: locate a skill, task, or occupation with confidence, reveal nearby related nodes, find paths between two nodes as learning journeys, and rank those paths with a named policy. Please wire it so the taxonomy data is treated as versioned suites from a separate taxonomy source, and keep the code organized around assistant logic, skills, typed result contracts, run logs, and a thin API layer.

Make it feel like a real working project, with a clean Python package, tests, and a simple FastAPI entry point if that fits. Use the docs in the repo as the source of truth, and look up current docs online if you need to. I’d also like the basic developer setup to work from the README, including environment config and running tests.