Google's TimesFM is a 200M parameter foundation model that does zero-shot forecasting on any univariate time series without training. Feed it sales, sensor data, weather readings, or vitals and get point forecasts plus calibrated prediction intervals (10th to 90th percentile). The skill includes a preflight checker that verifies you have the 4GB RAM and disk space before downloading the 800MB model, which matters because the agent won't crash your machine mid-run. Works CPU-only if needed, faster on GPU. Good for when you want probabilistic forecasts without tuning ARIMA parameters by hand or need to batch hundreds of series at once. Handles up to 16,384 context points per series.
npx skills add https://github.com/k-dense-ai/scientific-agent-skills --skill timesfm-forecasting