Build a local browser dashboard for digging into AI training rollouts — the text the model actually produced during training. Works on any training output (RL, SFT, DPO, GRPO, eval suites, custom research code, Tinker, prime-rl, verifiers, HuggingFace, anything else) by detecting where the rollouts live and rendering them as a browsable, click-to-read interface. Use this skill whenever the user wants to inspect a training run, debug model outputs, browse generations, see what the model wrote, or "dig into rollouts" — even if they don't say "dashboard". Rollouts are the focus; metrics and configs are secondary context.
2026-06-17