What problem does it solve? Forecasting time series traditionally requires training and tuning custom models like ARIMA or ETS for each dataset. This Skill provides zero-shot forecasting with Google's TimesFM foundation model, returning point forecasts with calibrated prediction intervals for any univariate series without training, while a mandatory preflight checker prevents out-of-memory crashes before the model loads. ## Core Features & Use Cases - Zero-Shot Forecasting: Feed any univariate series (sales, sensor readings, prices, vitals, weather) and receive point forecasts plus 10-quantile prediction intervals, with context lengths up to 16,384 points. - Covariate Forecasting (XReg): Incorporate dynamic numerical, dynamic categorical, and static exogenous variables via forecast_with_covariates() using TimesFM 2.5 and the timesfm[xreg] extra. - System & Dataset Preflight: scripts/check_system.py verifies RAM, GPU/VRAM, disk space, and Python version, and estimates memory for your specific dataset size before loading the ~800 MB model. - CSV Forecasting CLI: scripts/forecast_csv.py runs an end-to-end pipeline from CSV input to forecast output in CSV or JSON with 80% and 90% intervals. - Use Case: Forecast 52 weeks of demand from a weekly_demand.csv file, then evaluate accuracy with MAE, RMSE, MAPE, and prediction-interval coverage, or detect anomalies by flagging values outside the 90% quantile band. ## Quick Start Run python scripts/check_system.py to verify your machine, then ask the agent to forecast the value column of your CSV file 24 steps ahead with prediction intervals using TimesFM 2.5.
