Skip to main content
A multi-week migration is not one task. Split it into reviewable units. Each unit produces a concrete result a person checks before the next one starts. This keeps plans reviewable, limits risk and gives the next task verified context.
1

Inventory the current implementation

Explore the codebase, trace behavior and list the modules a migration would touch. Start in Interactive Chat and narrow the scope with follow-up questions. See Interactive Chat vs autonomous workflow vs CLI.
2

Split the work into checkpoints

Make each unit one outcome with a clear boundary, relevant context, acceptance criteria and a way to verify it. See Interactive Chat vs autonomous workflow vs CLI.
3

Migrate one bounded path

Refactor or update one path while preserving existing behavior and your conventions. See Work with existing code.
4

Add or update verification

Generate tests for the path you changed and run them in your pipeline. See Generate tests for existing code and Test Suite pipeline stage.
5

Review before the next unit

A person inspects the plan and the code changes before the next checkpoint begins. See Review the execution plan and Review code changes.
6

Repeat

Each checkpoint can become its own Jira task and workflow run. See Set up an autonomous workflow with Jira.

Plan for recovery

Releases can be rolled back by restoring a known good deployment. See Rollback strategy.

Check the limits first

Before you plan monorepo or database migration work, read Known limitations.

Next

Add and maintain test coverage

Keep production reliable