ln-76-architecture-migration-planner

Start by establishing the current and target states from repository evidence, then map the gap across modules, data stores, APIs, and consumers. This Skill structures the work into preparation, coexistence, migration, cutover, stabilization, and removal phases, each with entry gates and rollback actions.

Use the expand/migrate/contract pattern: apply additive schema changes first, backfill data with batched idempotent jobs, then remove old structures only after verified zero use. The Skill keeps additive and destructive operations in separate releases with explicit approval gates.

No. The Skill is strictly read-only except for writing the approved migration plan document. It does not execute migrations, edit product code, run deployments, or change task trackers; execution is left to downstream implementation planning.

The Skill returns a BLOCKED verdict rather than inventing either state. It requires specific enough evidence to compute a gap, and marks unknown consumers or unverifiable claims as migration risks instead of assuming zero usage.

Avoid it for generic task planning, delivery review, or direct migration execution, which are explicitly out of scope. It fits only current-to-target architecture transitions where reversibility, compatibility, and data safety must be planned before implementation.