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. Read

   data_sources.yaml

   to check if

   {name}

   exists as a registered source
2. If not found, try fuzzy matching (case-insensitive, partial match)
3. If still not found, list available datasets and ask user to choose

Step 2: Validate the data brain exists

1. Check that

   .knowledge/datasets/{name}/manifest.yaml

   exists
2. If it doesn't exist, suggest: "Dataset '{name}' is registered but has no data brain. Run

   /connect-data

   to set it up."

Step 3: Update the active pointer

1. Read

   .knowledge/active.yaml

2. Update

   active_dataset

   to

   {name}

3. Append to

   switch_history

   (cap at 20 entries, FIFO)
4. Write updated

   .knowledge/active.yaml

Step 4: Confirm the switch

Read the target dataset's

manifest.yaml

and display:

Switched to: {display_name}
Tables: {table_count}
Date range: {date_range}
Connection: {connection.type} ({connection.database}.{connection.schema})
Last analysis: {last_used or "none"}
Metrics defined: {count from metrics/index.yaml or 0}

Anti-Patterns

1. Never silently switch — always confirm with a summary
2. Never switch mid-analysis — if working/ has artifacts from the previous dataset, warn: "You have in-progress work for {old_dataset}. Switch anyway?"
3. Never infer the dataset — only switch when explicitly requested via this skill