SRE

Overview

The SRE ensures that what you build survives production. It designs observability, defines error budgets, plans for failure, and makes sure that when things break — and they will — recovery is fast and well-documented. Its core question: "will this survive production?"

When It's Used

Invoked during /dr-design for reliability requirements, /dr-qa for load and resilience review, and /dr-archive Step 0.5 for postmortem analysis. In Consilium, it speaks as the voice of reliability.

Capabilities

• SLO/SLA definition — establishes service level objectives and error budget management
• Observability design — metrics (RED, USE, 4 golden signals), structured logging, distributed tracing
• Alerting strategy — defines what to page on, what to log, what to ignore to reduce alert fatigue
• Capacity planning — scaling assessment and resource forecasting
• Incident response — runbooks, escalation paths, communication templates
• Chaos engineering — "what if this service dies? what if this dependency is slow?"
• Postmortem facilitation — blameless failure analysis with actionable follow-ups
• Graceful degradation — circuit breakers, bulkheads, retry with backoff, fallbacks
• Deployment safety — canary releases, feature flags, rollback procedures

How It Works

The SRE reads the task definition, system patterns, and tech context, then evaluates the system through a reliability lens. During design, it defines SLOs and identifies failure modes. During QA, it stress-tests the architecture mentally — what happens when a dependency is down, when traffic spikes 10x, when a deploy goes wrong. After incidents, it facilitates blameless postmortems with concrete follow-up actions.

Example

/dr-design "Deploy new API service"
→ SRE defines SLOs: 99.9% availability, p99 latency < 200ms
→ Error budget: 43 minutes downtime per month
→ Observability: RED metrics + structured JSON logging
→ Alerting: page on error rate > 5%, log on p99 > 500ms
→ Degradation: circuit breaker on payment service dependency
→ Runbook: 3-step rollback procedure documented

Context Loading

Reads datarim/tasks.md, datarim/systemPatterns.md, and datarim/techContext.md. Applies datarim-system and performance skills on every invocation. Loads security skill for security-related reliability concerns.

Skills Used

datarim-system (always), performance (always), security (when needed).