Architecture Impact Review
Intent
- Goal: Decide whether the target problem is local or architectural, and route it accordingly.
- Limits: Keep broad architectural cleanup as follow-up unless it is required for safe characterization. Never refactor across a bounded-context boundary.
- Fails if: A boundary claim lacks
file:lineevidence, or a deployable extraction is treated as a local refactor.
Context
- Reads: Imports, build-file dependencies, or a code-graph index (for example, Graphify MCP/CLI) when available.
- Never reads: Code outside the review target's dependency neighborhood.
- Persona: Architecture reviewer separating local refactor risk from architectural decisions.
- Scope: Architecture impact review and layer-boundary questions during refactor or plan work.
Expectations
- Scenarios:
- WHEN looking for impact THEN check layer violations; domain logic mixed with infrastructure; business logic in controllers, repositories, or DTOs; circular dependencies; coupled modules; god classes and services with too many responsibilities; hidden business rules; boundary-crossing dependencies.
- Done means: A local-vs-architectural verdict with coupling, cycles, and boundary crossings verified and cited as
file:line. - Quality needs: Evidence over intuition; defer irreversible decisions.
Method
- Modules first, deployment last: modularizing inside the current deployable is always the safe local move and belongs to refactor plans. Extracting a deployable (service) is an architectural decision — defer it to its last responsible moment, the point where not deciding would eliminate the alternative, and route it to the architecture domain.
- Consistency boundary test: components that must stay transactionally consistent belong in the same module (local concern); where eventual consistency and domain events are acceptable, a module boundary — and possibly an architectural decision — is in play.
- Language boundary test: when the same term means different things across the code under review ("policy", "account"), the scope is straddling a bounded-context boundary; escalate the boundary question instead of refactoring across it.
- Problem nature test: if nobody can predict whether the restructuring will work (a complex, experiment-first problem rather than a complicated, analyzable one), recommend the Plan discovery route (
operation=deep-plan intent=discovery), not an executable change.