switch-dataset

Skill: Switch Dataset

Purpose

Change the active dataset. Updates the active pointer, validates the target dataset exists, and confirms with a summary of what's now active.

When to Use

Invoke as /switch-dataset {name} when the user wants to analyze a different dataset than the currently active one.

Instructions

Step 1: Validate the target dataset

1. List available datasets by checking .knowledge/datasets/ directory for subdirectories with manifest.yaml files
2. Normalize the target name to lowercase for comparison (handles SALES-DATA → sales-data)
3. If {name} matches exactly (case-insensitive), proceed to Step 2
4. If not found, try fuzzy matching:
   • Check if {name} is a substring of any dataset name (e.g., "marketing" would match "marketing-prod")
   • Case-insensitive partial match
   • If exactly one match found, use that dataset and inform user: "Matched '{name}' to '{actual_dataset_name}'"
   • If multiple matches found, list all matches and ask user to choose
5. If still not found, list all available datasets with brief descriptions (from manifests) and ask user to choose or suggest running /connect-data to add a new dataset

Step 2: Check if already active

1. Read .knowledge/active.yaml to get the current active_dataset
2. If {name} matches the current active dataset (case-insensitive):
   • Inform the user: "The {name} dataset is already active."
   • Display the current dataset summary (same format as Step 6)
   • List other available datasets they could switch to instead
   • STOP here (do not proceed to Step 3)

Step 3: Validate the data brain exists

1. Check that .knowledge/datasets/{name}/manifest.yaml exists
2. If it doesn't exist, suggest: "Dataset '{name}' directory exists but has no manifest. Run /connect-data to set it up."
3. If manifest is missing, STOP (do not proceed)

Step 4: Check for in-progress work (CRITICAL SAFETY CHECK)

This step prevents accidental loss of analytical work. When you switch datasets, files in working/ become contextually orphaned — they contain SQL queries, charts, and data tied to the OLD dataset's schema and tables, which won't match the NEW dataset's structure. This doesn't delete the files, but it makes resuming that work much harder because the context has changed.