datasets

Skill: Datasets

Purpose

List all connected datasets with their status, table counts, and last analysis date. Also owns the switch procedure: /switch-dataset changes which dataset is active.

When to Use

Invoke as /datasets when the user wants to see what datasets are available, and as /switch-dataset {name} (or "switch to the other dataset") to change the active dataset.

Instructions

Step 1: Discover available datasets

The system supports two discovery paths:

Path A: Registry-first (preferred)

• Read data_sources.yaml to get the official list of registered sources
• If the file exists and has entries, use this as your source of truth

Path B: Brain-first (fallback when registry is empty)

• If data_sources.yaml is empty or missing, scan .knowledge/datasets/ directory
• Each subdirectory represents a dataset (directory name = dataset ID)
• Read each dataset's manifest.yaml to get connection details and metadata

Use whichever path yields results. Many installations have datasets in .knowledge/datasets/ but an empty data_sources.yaml registry — this is normal during initial setup or when datasets are added manually.

Step 2: Read the active pointer

Read .knowledge/active.yaml to determine which dataset is currently active.

Step 3: Enrich with manifest data

For each discovered dataset (whether from registry or directory scan), read .knowledge/datasets/{name}/manifest.yaml to get:

• display_name — human-readable name
• connection.type — connection type (csv, local_duckdb, snowflake, postgres, bigquery)
• connection.database or other connection-specific fields
• last_profiled: when the dataset was last profiled (null until data-profiling runs)

Table counts, date ranges, and row counts come from the dataset's schema.md or last_profile.md when those exist; connect-data's registration writes none of them. If a manifest is missing or those files have not been written yet, show what you can determine from the directory structure and note that the dataset needs profiling.