Snowflake Migration Skill¶

The Snowflake Migration Skill is an AI-powered skill for Cortex Code that guides you through an end-to-end database migration to Snowflake. It provides a conversational, interactive workflow — from connecting to your source database through code conversion, deployment, data migration, and validation.

Migrations take time. A full migration — especially for large workloads with hundreds of objects — is not completed in a single session. The skill automatically tracks your progress at every step. Each time you start a session, it reads your project state and picks up exactly where you left off. There is no need to start over.

Why use the Migration Skill¶

Cortex Code on its own is a capable coding agent, but a database migration involves coordination across dozens of tools, stages, and hundreds of objects over days or weeks. The Migration Skill adds the structure, automation, and domain expertise that a general-purpose agent doesn't have.

What the skill gives you¶


Benefit	What it means for you
Guided end-to-end workflow	You don't need to know the right sequence. The skill moves through connection, extraction, conversion, assessment, deployment, data migration, and validation automatically.
Session persistence	Close your terminal and come back tomorrow. The skill picks up exactly where you left off — no repeated setup, no lost progress.
SnowConvert integration	Source SQL is translated deterministically by SnowConvert before AI touches it. You start from a high-quality baseline, not a best-effort LLM rewrite.
Dependency-aware deployment	The skill analyzes object dependencies and builds deployment waves so objects are deployed in the right order. You don't manually sort hundreds of tables and views.
Two-sided testing	Functions and procedures are tested against source-side baselines automatically. Failures trigger a fix loop — the agent diagnoses, patches, and re-tests until the output matches.
Reusable fix rules	Every correction you make can be extracted into a rule and propagated across the entire project. The skill gets smarter as you go.
Zero setup	All dependencies (Python packages, SnowConvert AI, ODBC drivers) are installed automatically the first time the skill runs.

Skill vs. plain Cortex Code¶


	Migration Skill	Plain Cortex Code
Structured multi-stage workflow	Yes — six stages from connect to migrate	No — you drive every step manually
Automatic state tracking	Yes — resumes across sessions	No — you restart context each session
SnowConvert deterministic conversion	Yes — integrated	No — you run it yourself and import results
Source database connectivity	Yes — connects, extracts, and migrates data	No — no built-in database connectors
Deployment wave planning	Yes — dependency analysis and interactive wave editor	No — you plan deployment order manually
Automated testing loop	Yes — baseline capture, two-sided validation, auto-fix	No — you write and run tests yourself
Reusable fix rules	Yes — extract, search, apply, propagate	No — fixes are one-off

How people use the skill¶

There are three common ways to start a migration. The skill adapts to each.

Start a new migration (greenfield)¶

You have a source database and want to migrate it to Snowflake from scratch. The skill walks you through every stage: connect to the source, extract objects, convert code, assess the workload, deploy, migrate data, and validate.

Use the migration-guide skill to migrate a database

Resume an in-progress migration¶

You already started a migration in a previous session — maybe days or weeks ago. The skill reads your project state and picks up at the exact point where you stopped. No re-extraction, no re-conversion.

Continue

Import an existing migration¶

You already have .sql files from a previous SnowConvert run or another source. The skill imports them into a project and picks up at the conversion or deployment stage.

Import SQL files from ./my-exported-scripts/

What you can do¶

The skill guides you through the full migration lifecycle and lets you jump directly to any stage at any time. You can pause and resume across sessions, and multiple users can collaborate on the same project — the skill tracks which code units each person is working on so effort isn't duplicated.


Capability	Description
Connect to source	Set up a connection to your source database with credentials stored securely for reuse
Extract source code	Pull DDL and stored procedures directly from a live database, or import local `.sql` files
Convert code	Translate source SQL to Snowflake-compatible SQL via SnowConvert, with a full EWI report
AI-assisted conversion	AI explains remaining conversion issues, suggests fixes, and applies them interactively
Assess workloads	Generate an interactive HTML report covering deployment waves, object exclusions, dynamic SQL patterns, and SSIS/Informatica ETL analysis
Deploy objects	Deploy converted tables, views, functions, and procedures to Snowflake wave by wave
Migrate data	Copy rows from source tables to Snowflake with automatic row-count validation
Test functions and procedures	Capture source-side baselines and run two-sided validation to confirm output equivalence
Fix and rule engine	Create reusable fix rules from corrections you make and propagate them across the project automatically
Convert ETL pipelines	Translate SSIS packages and Informatica workflows to dbt, using deterministic conversion with optional AI-assisted remediation
Validate data	Compare row counts and data between source and Snowflake tables after migration to confirm completeness
Repoint reports	Repoint Power BI reports to use Snowflake as the data source

Supported source systems¶

Not all capabilities are available for all source systems. The following table shows what is supported today and what is coming soon.


Capability	SQL Server	Redshift	Teradata	Oracle	Other source systems
Code extraction	Yes	Yes	Planned	Planned	—
Deterministic code conversion	Yes	Yes	Yes	Yes	Yes
AI conversion and verification	Yes	Yes	—	—	—
Code deploy	Yes	Yes	Planned	Planned	—
SSIS to dbt (deterministic)	Yes	Yes	Yes	Yes	Yes
Informatica to dbt (deterministic)	Yes	Yes	Yes	Yes	Yes
AI conversion of SSIS to dbt	Yes	Yes	Yes	Yes	Yes
AI conversion of Informatica to dbt	Yes	Yes	Yes	Yes	Yes
Cloud data migration	Yes	Yes	Planned	Planned	—
Cloud data validation	Yes	Yes	Yes	Planned	—
Testing framework	Yes	Yes	Planned	Planned	—
Testing with synthetic data	—	—	Planned	Planned	—
AI assessment (via Cortex Code)	Yes	Yes	Yes	Planned	—
Power BI report repointing	Yes	Yes	Yes	Yes	Synapse and PostgreSQL only

Other dialects with deterministic conversion support: Azure Synapse, Sybase IQ, Google BigQuery, Greenplum, Netezza, PostgreSQL, Spark SQL, Databricks SQL, Vertica, Hive, IBM DB2

Available extraction scripts: Teradata, SQL Server, Synapse, Oracle, Redshift, Netezza, Vertica, DB2, Hive, BigQuery, Databricks, Sybase IQ

Prerequisites¶

Before starting a migration, ensure you have:

Cortex Code installed (the migration skill is bundled).
A Snowflake account with a connection configured in ~/.snowflake/connections.toml.
A source database (SQL Server or Redshift) accessible from your machine.

All other dependencies (uv, scai, Python packages) are installed automatically when the skill runs for the first time.

Quick Start¶

1. Launch Cortex Code:

cortex

2. Start a migration:

Use the migration-guide skill to migrate a database

The migration-guide skill activates automatically, confirms plugin installation with you, registers the plugin, and walks you through the full migration workflow. If a project already exists, the agent picks up where you left off.

Migration workflow¶

The skill guides you through six stages. Each session starts by detecting your current progress and resuming automatically.


Stage	Name	What happens
1	Connect	Set up a connection to your source database
2	Init	Create a local migration project
3	Register	Extract DDL and code from the source, or import local `.sql` files
4	Convert	Translate source SQL to Snowflake-compatible SQL via SnowConvert
5	Assess	Generate an interactive report covering waves, exclusions, dynamic SQL, and ETL
6	Migrate	Deploy objects, migrate data, validate output, and fix errors — wave by wave

Stage 6 is where the bulk of the work happens. AI-driven testing, conversion remediation, and iterative fix loops run here — often across multiple sessions and days. This is also the stage where collaboration pays off most: multiple users can work on the same project simultaneously, each picking up different code units while the skill coordinates to prevent duplicated effort.

Every session shows a live progress checklist so you always know where you stand:

✅  1. Connect             — Connected to SQL Server
✅  2. Init                — Project initialized
✅  3. Register            — 342 objects registered
◐   4. Initial Conv        — 280/342 converted
⬚   5. Assess              — Not run
⬚   6. Migrate Objects     — 0/120 tables deployed

Skills Reference¶

The skill is organized as a skill tree. The root skill detects your project state and delegates to the right sub-skill. You can also invoke any skill directly by describing what you want.

Setup Skills (Stages 1–5)¶

connection¶

Walks you through connecting to your source database. The agent collects credentials, tests the connection, and saves it for reuse across sessions. Supports:

SQL Server — configures ODBC driver, host, port, and authentication.
Amazon Redshift — configures host, port, database, and IAM or password authentication.

register-code-units¶

Gets source code into the migration project. Two paths are available:


Path	When to use
Extract from database	You have a live source connection and want the agent to pull DDL and object code directly
Import local files	You already have `.sql` files on disk and want to import them into the project

convert¶

Runs SnowConvert to translate your source SQL (T-SQL or Redshift SQL) into Snowflake-compatible SQL. After conversion, the agent presents:

Total objects converted successfully.
EWI (Early Warning Issue) summary broken down by severity (errors, warnings, informational).
A list of objects that require manual review.

assessment¶

Generates an interactive multi-tab HTML report. The assessment includes four analyses that can be run individually or together:


Analysis	What it does
Deployment Waves	Analyzes object dependencies to produce an ordered deployment sequence. Objects within a wave have no inter-dependencies; waves are ordered so dependencies are always deployed first.
Object Exclusion	Identifies objects that do not need migration: temporary tables, staging objects, deprecated objects, and test artifacts. Reduces scope before deployment.
Dynamic SQL Analysis	Classifies and scores Dynamic SQL patterns in your converted code. Identifies patterns that Snowflake handles natively, patterns requiring manual rewrite, and patterns with elevated migration complexity.
ETL/SSIS Assessment	Analyzes SSIS packages individually: classifies each package (Ingestion, Transformation, Export, Orchestration, Hybrid), maps control and data flow, and estimates migration effort.

The report is generated as a single self-contained HTML file. You can iterate on the wave plan interactively — for example, reprioritizing objects, adjusting wave sizes, or relocating specific objects — before locking it for deployment.

Migration Skills (Stage 6)¶

migrate-objects¶

The main deploy loop. Processes all objects in the current wave in dependency order:


Object type	What happens
Tables	Deployed to Snowflake, then data is migrated from the source.
Views	Deployed to Snowflake. Blocked views retry after their dependent functions/procedures pass.
Functions & Procedures	Deployed, tested against source output, and fixed if tests fail. The loop repeats until tests pass or the user decides to skip.

After each wave completes, the agent automatically advances to the next wave.

baseline-capture¶

Captures the expected output of source stored procedures and functions for use as test baselines. Two approaches are supported:


Approach	When to use
Query Logs	You have CSV logs of real `EXEC` or `CALL` statements from your source system. The agent parses these to extract parameters and expected outputs.
AI-Assisted	No logs are available. A swarm of specialized agents generates test cases covering business logic, data-driven scenarios, and edge cases by analyzing the source SQL.

Baselines are stored locally and uploaded to Snowflake so they can be used for two-sided validation (source output vs. Snowflake output) during the migrate-objects loop.

rule-engine¶

Manages reusable migration rules stored in Snowflake. Rules encode known source-to-Snowflake fix patterns and are shared across all objects in the project. Each rule can operate in two modes:


Mode	How it works
Regex	A regex find-and-replace applied mechanically to SQL files
AI-guided	The rule provides context and strategy; the AI interprets and applies it

The rule engine has four sub-capabilities:


Sub-skill	What it does
search	Scans a SQL file against all rules using regex pattern matching and Cortex semantic search. Returns matched rules ranked by relevance.
apply	Applies matched rules to local SQL files. Regex rules are applied automatically; AI-mode rules are shown for review before applying. Supports single-file and batch application.
extract	Creates a new reusable rule from a fix you just made. Works from an interactive before/after comparison or retroactively from git history.
propagate	Given a rule, finds every code unit in the project it applies to (via reverse regex + semantic search), then hands off to batch apply.

Rules accumulate over the lifetime of the project. Every time the agent fixes an object and extracts a rule, that rule becomes available to all subsequent objects — reducing manual effort as the migration progresses.

What You Can Ask¶

You do not need to follow the prescribed path. You can ask for any capability at any time.

Status and navigation¶


Prompt	What happens
`"What is the current state?"`	Shows the progress checklist
`"What should I work on next?"`	Returns the next dependency-ready object
`"Continue"`	Picks up the prescribed migration path

Setup¶

"Connect to my SQL Server database"
"Extract objects from the source"
"Import SQL files from ./my-scripts/"
"Convert my source code"

Assessment¶

"Run a full assessment"
"Generate deployment waves"
"I want a maximum of 30 objects per wave"
"Prioritize all Payroll objects in Wave 1"
"Identify temporary and staging objects"
"Analyze dynamic SQL patterns"
"Assess my SSIS packages"

Migration¶

"Deploy tables"
"Migrate data"
"Deploy and test the next function"
"Capture baselines for dbo.GetCustomerOrders"

Rule engine¶

"Search rules for this file"
"Apply all matched rules"
"Extract a rule from my last fix"
"Propagate this rule across the project"
"Show me all rules"

Example Workload¶

You can use AdventureWorksDW (https://learn.microsoft.com/en-us/sql/samples/adventureworks-install-configure) as an example source database to try the skill end-to-end. Substitute any SQL Server or Redshift database you have access to — the skill adapts to whatever source you connect.

Troubleshooting¶


Problem	Resolution
`migration-guide` skill doesn't appear	Ensure you're on the latest version of Cortex Code. Run `cortex --version` to confirm.
Plugin installation fails during skill setup	Check that `python3` is available on your `PATH`. The skill installs dependencies via Homebrew (macOS/Linux) or winget (Windows).
`scai` not found after skill install	Run the install hook manually: `migration-plugin/hooks/install-dependencies`. Or install directly: `brew install --cask snowflakedb/snowconvert-ai/snowconvert-ai`.
Snowflake connection errors	Verify your connection in `~/.snowflake/connections.toml` and confirm the connection name matches what you provided during setup.
Agent seems lost	Say `"What is the current state?"` — the agent re-reads project status and resets context.

Support¶

For help with the migration skill, contact: snowconvert-support@snowflake.com