eddytor-data-import

name	eddytor-data-import
description	Imports CSV data into Eddytor tables with schema inference, type mapping, and post-import validation. Activates for CSV import, data loading, schema preview, raw CSV content, type detection, or delimiter handling — even without the word "import."
license	CC-BY-NC-4.0
metadata	{"author":"eddytor","version":"1.0"}

Data Import

Default workflow: infer → import → constrain → validate

1. infer_schema with CSV content → review column types and domain candidates
2. import_csv with reviewed schema → creates table + inserts rows in one step
3. set_column_domain on domain candidates flagged by step 1
4. validate_domain_values + profile_table → confirm data quality

Always start with infer_schema. Skipping it means type surprises after import.

infer_schema — preview before committing

Returns row count, column names, Arrow types (with nullability), and domain candidates (low-cardinality string columns with distinct values listed). The delimiter is auto-detected internally for parsing but is not part of the response.

Review the output. Common adjustments:

• Dates as Utf8 → should be Date32/Timestamp
• IDs as Int64 → use Utf8 if they have leading zeros
• Boolean-like strings ("yes"/"no") → set a fixed domain after import

import_csv — one-step table creation

{
  "table_name": "products",
  "location": "abfss://[e-mail]/master-data",
  "primary_key_column": "product_id",
  "csv_content": "product_id,name,category,price\nP001,Widget,Electronics,29.99\n...",
  "non_nullable_columns": ["product_id", "name", "category"]
}

delimiter auto-detects , ; \t | — set explicitly only if auto-detection fails.

Post-import validation loop

1. profile_table → verify row count, check null distributions
2. validate_constraints → check expressions hold
3. validate_domain_values → catch typos in constrained columns
4. If issues: fix with merge_rows → re-validate

Gotchas

• table_name is bare only (products), never prefixed with catalog/schema.
• primary_key_column must be a column name that exists in the CSV.
• If import fails (type conflict, duplicate PKs), the table is not created — fix data and retry.
• csv_content has MCP message size limits (~5MB). For larger files: create the table with a first batch, then use insert_rows for the rest.
• Empty strings in CSV are not the same as NULL — check nullability settings.
• "001234" becomes 1234 if inferred as Int64 — ensure Utf8 type to preserve leading zeros.