| name | design-project-architecture |
| description | Assess, design, or evolve module, data, runtime, deployment, or team-ownership boundaries. Use when those boundaries or responsibility contracts are the task; skip isolated code changes, folder cleanup, and refactors that leave boundaries intact. |
Design Project Architecture
Choose one mode before proposing changes.
- Assessment: map current boundaries, evidence, risks, and gaps against stated drivers.
- Design: compare viable boundary options and define the smallest target that satisfies the drivers.
- Evolution: preserve behavior while moving toward a target through reversible, observable steps.
Establish the decision
- State the system scope, stakeholders, constraints, affected boundary types, and decision owner. Turn the relevant concerns into two to five measurable quality-attribute scenarios: source and stimulus, environment, affected asset, required response, metric, and acceptance threshold. Read architecture principles for the template.
- Map only the views needed to answer the decision: module dependencies and contracts; data ownership and flows; runtime calls and failure paths; deployment and operational topology; and team ownership and interactions. For an existing system, use existing-project evolution. For a new one, use new-project design.
- Stop before finalizing an irreversible contract, data, security, availability, deployment, or ownership decision if its drivers are unknown. Record the missing business, regulatory, load, failure-tolerance, rollback, operating-capacity, or accountable-owner information; present reversible options or experiments instead of selecting a target.
Shape boundaries from drivers
Give each responsibility and state transition an explicit owner and contract. Choose cohesion, information hiding, dependency direction, integration style, and operational scope by the change, consistency, performance, security, availability, cost, and team-flow drivers.
Treat dependency cycles, shared mutable state, replacement seams, and deployment splits as trade-offs rather than universal defects or goals. Retain a cycle or shared state only with a named benefit, a bounded scope, an owner, an enforcement mechanism, and an exit condition. Add an adapter or replacement seam only for a credible source of change, isolation, test, or migration value. Split runtime or deployment units only when selected drivers outweigh added latency, failure, observability, release, and consistency costs.
For every data or distributed boundary, resolve or explicitly mark out of scope: source of truth, ownership, schema evolution, transaction and consistency model, derived data, idempotency, retries, reconciliation, retention, and rollback. Account for a team's cognitive load, operational burden, and interaction modes. Time-box collaboration for discovery, define service contracts and support expectations for X-as-a-Service, and define an objective and exit condition for facilitation.
Verify proportionately
Turn each important scenario into the smallest credible fitness function. Specify its signal, threshold, cadence, owner, and failure action. Use a reviewable manual check for low-risk one-off decisions; use automated import, contract, schema, or deployment checks for recurring risks; use representative load, resilience, security, restore, or reconciliation checks and production signals when the consequence warrants them. Read architecture validation for examples and evidence expectations.
Record the resulting boundary map, selected scenarios, assumptions, trade-offs, owners, and consequential decisions. Use diagrams and ADRs only when they resolve a stakeholder question. Consult sources for primary guidance or contested choices.
Finish when the applicable boundaries are understandable, the selected option is justified by measurable drivers, the resulting risks have owners and evidence, and unverified assumptions or deferred decisions are explicit.
