Cloud & DevOps Workflows

Use when an approved system design exists and the product includes LLM, agent, retrieval, classifier, extractor, or model-driven automation components that need architectural definition before implementation. Produces capability classification, model contracts, context and memory strategy, retrieval and grounding topology, tool and action surface, agent control flow, failure taxonomy, evaluation and guardrail plans, cost and latency budgets, model routing, observability requirements, and implementation handoff notes. Do not use for generic backend service design, data pipeline design, product requirement definition, or vendor-specific SDK scaffolding; use backend-architecture, data-architecture, idea-development, or implementations/ai/ instead.

Use when an approved ai-architecture.md needs eval or observability coverage for an Anthropic Claude capability. Produces eval datasets, regression gates, prompt/model versioning, token/cost and cache telemetry, and traces. Not for model selection, RAG topology, or tool/agent design.

Use when an Anthropic Claude capability from an approved ai-architecture.md must make cachecontrol, cache TTL, extended-thinking budget, or long-context packing deliberate against a cost/latency budget. Produces a cache/context plan plus telemetry. Not for RAG, tool, or agent design.

Use when an approved ai-architecture.md defines an Anthropic Claude retrieval-augmented capability. Produces a retrieval adapter, context packing, grounding prompt, Citations-API sourcing, a hallucination gate, and retrieval evals. Not for tool execution, schema-only output, or agent design.

Use when an approved ai-architecture.md needs an Anthropic Claude capability returning schema-bound JSON, typed objects, classifications, or extractions. Produces Messages API wiring, schema validation, bounded repair, and tests. Not for RAG, tool execution, or agent control flow.

Use when an approved ai-architecture.md defines an Anthropic Claude capability whose model calls tools or takes side-effecting actions. Produces tool schemas, an auth-enforcing execution adapter, idempotency, audit logging, and failure tests. Not for schema-only output, RAG, or agent design.

Use when an approved ai-architecture.md defines a multi-agent workflow and AutoGen is the chosen framework. Produces the group-chat/team topology, role-to-agent mapping, critic-gated termination, budgets, and eval wiring. Not for single-agent runtimes, tool-surface design, or provider SDK work.

Use when an approved ai-architecture.md defines a tool surface and AutoGen is the chosen framework. Produces tool schemas, an authorization-enforcing execution adapter, idempotency, audit logging, and tool-failure tests. Not for agent topology, retrieval design, or provider SDK work.

Use when designing, reviewing, or hardening the AWS Organizations and account topology for a system after infrastructure-platform and security have decided the org structure and environment ladder. Produces the AWS Organizations OU structure, landing-zone approach (Control Tower or custom), Service Control Policy guardrails, environment-isolated account layout, central billing and cost-allocation tagging, and baseline AWS Config/audit posture. Do not use for in-account VPC/IAM/KMS design, workload compute selection, observability wiring, or DR topology; use the other aws archetype skills (network-and-identity-foundation, workload-runtime-and-deployment, observability-and-cost-readiness, dr-and-multi-region-readiness) instead.

Use when designing and rehearsing AWS disaster-recovery and multi-region posture for a workload after the runtime and observability exist and reliability and operations have decided RPO/RTO targets and failover ownership. Produces multi-AZ baseline, tier-driven multi-region topology (active-passive / active-active), cross-region data replication (RDS replicas, S3 CRR, DynamoDB Global Tables), Route 53 health-check failover, AWS Backup posture, and a documented, rehearsed failover drill with measured RPO/RTO. Do not use for org/account topology, network/identity foundation, single-region workload runtime, observability/cost wiring, or Terraform module/state mechanics; use the other aws (or Family H) skills.

Use when designing, reviewing, or hardening the in-account AWS network and identity foundation after account topology exists and security and infrastructure-platform have decided trust zones and the identity model. Produces VPC topology (per-env, per-tier, multi-AZ subnets), inter-account connectivity (Transit Gateway / peering / PrivateLink), IAM Identity Center federation, IAM role-assumption patterns and permission boundaries, KMS CMK strategy, Secrets Manager with rotation, and Route 53 zone strategy. Do not use for org/account/SCP topology, workload compute selection, observability/cost wiring, or DR/multi-region; use the other aws archetype skills. Emits IaC-ready definitions, not Terraform module/state mechanics.

Use when wiring AWS observability and cost readiness for a workload or account after the runtime exists and operations and performance have decided the observability substrate and cost posture. Produces CloudWatch Logs/Metrics/Alarms, AWS Distro for OpenTelemetry and X-Ray tracing, SLO dashboards, Cost Explorer + AWS Budgets + Cost Anomaly Detection, FinOps tagging discipline consuming the org tag policy, and savings-plan/RI posture. Do not use for org/account/tag-policy enforcement, network/identity, workload runtime, or DR/multi-region, or Terraform module/state mechanics; use the other aws (or Family H) skills.

Use when selecting the AWS compute primitive for a workload and designing its load balancing, autoscaling, and deployment mechanics after the network/identity foundation exists and infrastructure-platform and reliability have decided the runtime substrate and SLOs. Produces the compute-primitive selection per workload class (Lambda / Fargate / EKS / EC2-ASG / RDS-Aurora / DynamoDB), ALB/NLB posture, autoscaling configuration, and deployment mechanics (CodeDeploy, blue/green for ALB, rolling for ECS) with rollback. Do not use for org/account topology, VPC/IAM/KMS foundation, observability/cost wiring, DR/multi-region, in-cluster Kubernetes manifests, or Terraform module/state mechanics; use the other aws (or Family G / Family H) skills.

Use when an approved ai-architecture.md defines a multi-agent workflow and CrewAI is the chosen framework. Produces the crew topology, role-to-agent mapping, critic-gated termination, budgets, and eval wiring. Not for single-agent runtimes, tool-surface design, or provider SDK work.

Use when an approved ai-architecture.md defines CrewAI tasks or callable tools. Produces task decomposition, tool schemas, an auth-enforcing execution adapter, idempotency, audit logging, and failure tests. Not for crew topology, retrieval design, or provider SDK work.

Review or assess a data architecture, data model, or data platform design. Evaluates six data-specific quality attributes (DAMA-DMBOK aligned), topology choice, data contracts, GDPR/AI Act compliance, governance gaps, and data mesh readiness. TOGAF Phase C aware.

Use when containerizing a JVM service for production.

Use when adding background tasks or event integration to a scaffolded FastAPI service after the broker and contracts are declared in backend-architecture. Adds Celery/RQ/arq or Kafka wiring, a transactional outbox, idempotent consumers, retry/DLQ, and Testcontainers tests. Not for the shell, auth, observability, or performance.

Use when adding auth/authz, HTTP-surface hardening, or an OWASP review to a scaffolded FastAPI service after the auth provider decision is approved or deferred. Adds the authentication, default-deny authz, hardening, and security-test layer the scaffold defers. Not for observability, tasks, or performance.

Use when making a scaffolded FastAPI service production-observable after reliability SLO targets are approved or deferred. Replaces the telemetry seam with OpenTelemetry, prometheus-client RED metrics, trace-correlated structlog, and SLO/burn-rate alerts. Not for the shell, auth, tasks, or performance.

Use when hardening a scaffolded FastAPI service for load and failure after performance budgets and SLO targets are approved or deferred. Adds async-path discipline, the worker model, pool sizing, caching, breakers/bulkheads, retry budgets, and a CI load-test gate. Not for the shell, auth, observability, or tasks.

Use when scaffolding or production-hardening a FastAPI service shell after backend architecture is approved or deferred. Produces the settings/logging/health/error-handling/lifespan/DI/container baseline the other FastAPI skills extend. Not for auth, observability, tasks, or performance.

Use when creating, modernizing, or production-hardening a Flutter mobile application after mobile architecture is approved or intentionally deferred. Produces a production-ready app shell with flavor-separated entry points, environment and profile handling, layered error handling (runZonedGuarded, PlatformDispatcher, FlutterError, ErrorWidget), a structured logging client, crash reporting and performance tracing seams, the DI and session provider baseline downstream archetypes extend, and CI signing scaffolding without committing credentials. Do not use for navigation topology or deep-link wiring, state or data-fetching implementation, design-system or accessibility integration, or performance budget enforcement; use the other Flutter archetype skills instead.

Use when integrating a design system, theming, the component library, accessibility posture, the internationalization seam, or permission-request UX into a Flutter application after the app scaffold exists and mobile architecture is approved or intentionally deferred. Produces the token and theming layer extending the scaffold's ThemeData placeholder, a composed component library, the accessibility posture from mobile-architecture.md (screen reader, dynamic text, reduced motion, contrast, touch targets, RTL), an i18n seam, accessible auth UIs, and permission-request UX (rationale, denied, and blocked states). Do not use for app scaffolding, navigation topology or layout placement, state management or token logic, the native permission engine, or performance budget enforcement; use the other Flutter archetype skills instead.