aws-waf-reliability-skill

When to Use

Apply this skill at design time, when the solution-architect agent has produced a design.md and per-component contracts and is composing the per-pillar review block. The implementation agent re-applies it before transitioning a contract from accepted to implemented. Do not apply this skill to runtime incident triage, live failover execution, or chaos-experiment runs — its scope is design-time Reliability assessment.

Trigger conditions:

• The agent is writing the Reliability review block of design.md for a new or revised feature.
• The agent is reviewing a draft contract and needs the Reliability checklist applied.
• The agent is verifying recovery objectives (RTO and RPO) are named, measurable, and consistent with the chosen DR strategy.
• The agent is checking multi-AZ posture, failover paths, or read-replica configuration on stateful components.
• The agent is verifying retry, timeout, idempotency, and circuit-breaker policies on every cross-service call.
• The agent is auditing dependency mapping, quota headroom, and failure-mode analysis for the feature.

Procedure

1. Walk the design questions. Apply the questions in references/rel-design-questions.md to the artefact under review. A missing answer is a finding, not a passing default.
2. Scan for anti-patterns. Match the design against the catalogue in references/rel-antipatterns.md. Each match is recorded with severity and the specific evidence (file path + section).
3. Run the checklist. Walk references/rel-checklist.md end-to-end; every item is satisfied with evidence, deferred with rationale, or flagged as a gap.
4. Verify recovery objectives. For every component holding state or serving a critical path, confirm RTO and RPO are named with units (minutes/hours, bytes/transactions). The chosen DR strategy (backup-restore, pilot light, warm standby, multi-site active-active) must be consistent with those numbers; mismatches are HIGH gaps.
5. Verify multi-AZ and failover. For every stateful or singleton component, confirm AZ posture is named (single-AZ with rationale, multi-AZ active-passive, multi-AZ active-active) and the failover trigger and detection mechanism are documented.
6. Verify retry, timeout, idempotency. For every cross-service call (HTTP, SQS, SNS, EventBridge, Step Functions, downstream SaaS), confirm a timeout, a retry policy with bounded attempts and jitter, and an idempotency strategy. Unbounded retries, naked timeouts, and non-idempotent retried writes are HIGH gaps.
7. Emit the Reliability review block. Write the block into design.md using assets/rel-review.md.tmpl as the structure: pillar header, design-question summary, anti-patterns matched, FMA outcome, checklist outcome, and explicit gap list with severity.