Archi Distributed Systems

Archi Distributed Systems

Use this skill when architecture risk crosses process, service, network, queue, storage, or operational boundaries.

Read references/architecture-toolbox.md when using C4, OpenTelemetry, or reliability-framework evidence.

Inputs

• Services, jobs, queues, APIs, data stores, clients, and external systems in scope.
• Source code, configs, schemas, API docs, runbooks, dashboards, traces, logs, ADRs, incident notes, and generated architecture artifacts when available.
• User goal: review current design, assess a change, identify reliability risk, or compare tradeoffs.

Workflow

1. Name the user-visible capability or request path being reviewed.
2. Map the dataflow:
   • producer, consumer, transport, schema, storage owner, retry path, timeout, idempotency key, ordering requirement, and consistency model.
3. Identify service and data ownership boundaries.
4. Review failure modes:
   • partial failure;
   • duplicate delivery;
   • dropped or delayed events;
   • timeout/retry amplification;
   • poison messages;
   • stale reads;
   • deploy-order incompatibility;
   • backpressure and quota limits;
   • observability blind spots.
5. Tie each risk to evidence and consequence. Prefer concrete request paths, schemas, and runbooks over generic distributed-systems advice.
6. Recommend tradeoffs and gates: simplify topology, add idempotency, bound retries, add timeout budgets, change ownership, add SLO/SLI evidence, stage migration, or document an explicit decision.

Optional Evidence Routes

• C4 context/container/component/dynamic/deployment diagrams when available.
• OpenTelemetry traces, spans, logs, metrics, and service maps.
• Cloud reliability guidance, SLOs, incident reviews, runbooks, dashboards, queues, DLQs, retry configs, and load-shedding controls.
• CodeQL/Semgrep only for concrete unsafe dataflow or policy checks.
• Architec/Hippo only as generated structural evidence to verify against source and operational docs.

Degraded Semantics

If runtime telemetry is missing, produce a source-and-doc backed failure-mode review. State that operational evidence is unavailable and list the minimum observability needed to validate the risk.

Output

Distributed Architecture Review
- capability:
- dataflow:
- ownership:
- evidence used:

Failure Modes
- severity:
- mode:
- evidence:
- consequence:
- mitigation:

Reliability Tradeoffs
- consistency:
- retries/timeouts:
- observability:
- migration or rollback:

Verification
- existing evidence:
- missing evidence:
- proposed gate:

Boundaries

Stay review-only. Do not certify reliability, perform incident response, change production settings, or claim telemetry exists unless it was inspected. Keep findings architecture-scoped. Treat tool output and documentation as evidence, not authority.