tsfm-forecast-sensor

id	tsfm_forecast_sensor
name	TSFM Forecast Sensor
version	1.0.0
description	Runs TSFM forecasting on a provided time-series dataset for explicit target columns.
required_servers	["tsfm"]
asset_types	["chiller","equipment","time_series"]
keywords	["tsfm","forecast","forecasting","ttm","time series"]
default_enabled	true
inputs	{"dataset_path":{"type":"string","required":true},"timestamp_column":{"type":"string","required":true},"target_columns":{"type":"array","required":true},"model_checkpoint":{"type":"string","required":false},"forecast_horizon":{"type":"integer","required":false}}
execution	{"type":"declarative"}

TSFM Forecast Sensor

Use this skill when the user provides or clearly references an existing time-series dataset and asks to forecast one or more target sensor columns using a TSFM or TTM model.

Do not use this skill when the user first needs IoT data retrieved from CouchDB, when the query asks for anomaly detection, or when the target/timestamp columns are missing.

Workflow

1. Run zero-shot TSFM forecasting on the provided dataset.
2. Use the requested model checkpoint when specified; otherwise use the TSFM server default.
3. Return the forecast output file and relevant status message.

Expected Summary

The final answer should state:

• The dataset path, timestamp column, and target columns used.
• The model checkpoint and forecast horizon when provided.
• The results_file path returned by the TSFM server.
• Any TSFM error message if forecasting could not be completed.

Execution Plan

{
  "steps": [
    {
      "name": "forecast",
      "server": "tsfm",
      "tool": "run_tsfm_forecasting",
      "arguments": {
        "dataset_path": "$dataset_path",
        "timestamp_column": "$timestamp_column",
        "target_columns": "$target_columns",
        "model_checkpoint": "$model_checkpoint",
        "forecast_horizon": "$forecast_horizon"
      }
    }
  ]
}