Tensorboard

This pulls in TensorBoard for visualizing your ML training runs directly from Claude. You get the standard toolkit: loss curves, accuracy plots, model graph visualization, embedding projections, and performance profiling. It's the same TensorBoard you'd use anywhere else (27k+ GitHub stars, 20M+ downloads yearly), just wired up as a skill so you can spin it up without leaving your conversation. Honestly most useful when you're iterating on experiments and need to compare runs side by side or debug why your model isn't converging. The skill itself is pretty straightforward, more of a convenience wrapper than anything novel.