Phoenix Cli

This wraps the Phoenix CLI for tracing and debugging LLM applications. The interesting bit is the structured workflow: open coding lets you sample traces and write freeform notes about failures, then axial coding groups those notes into a MECE taxonomy with counts so you know which evals to write. Every annotation gets tagged with a coding identifier so you can query or revert a whole analysis session as a unit. The skill works at trace, span, or session level depending on where your failure modes actually live. It also handles profiles for switching between local and cloud Phoenix instances, plus the usual list/get/annotate commands you'd expect from a resource CLI. If you've instrumented an app and are staring at a pile of traces wondering what's actually breaking, this gives you a method.