Refactoring a brittle Ruby import script into a robust Sidekiq job is solid AI territory.

This is a one-time refactor with a well-understood pattern: wrap a script in a Sidekiq worker, add idempotency, validation, and tests. The structure is highly repeatable across similar legacy-script modernization tasks.

The high-level deliverables are crisp (Sidekiq job, rollback, logging, tests), but edge-case business rules — how to handle duplicate emails, what phone formats are valid, what 'resumable' means for this specific schema — require clarification or assumptions that a human should validate.

The agent needs the existing 400-line script, the PostgreSQL schema, and ideally sample CSV files with edge cases. Without these, the agent must make assumptions. If provided, the task becomes highly executable.

Bugs in the refactored job could corrupt customer data or silently drop records in production, which is serious. However, the task is code generation for review — a human should run tests and review before deploying, keeping actual error cost manageable.

Decisions about duplicate resolution strategy, rollback granularity, and validation thresholds encode business logic that only the team knows. An agent can make reasonable defaults, but a human must confirm these choices before the code ships.