timesfm-forecasting

SUMMARY

What it does

This Agent Skill provides zero-shot time series forecasting using Google's TimesFM foundation model. It is designed for univariate time series data (e.g., sales, sensors, energy, vitals, weather) and requires no custom model training. The skill supports inputs as CSV files, DataFrames, or arrays, and outputs point forecasts along with prediction intervals. It includes a mandatory preflight system checker script that verifies RAM, GPU memory, and disk space before model loading to prevent system crashes. The skill wraps TimesFM for local inference and includes scripts for system checking and CSV forecasting. The source documentation describes usage patterns, installation steps, and common pitfalls. The skill is distributed under the MIT license (as declared in the submitted record) and is intended for agent-friendly local inference.