SRE Reliability Ops: Alerts, Budgets, Incidents, Capacity

Your pager goes off for CPU spikes nobody needs to act on, while the outage customers actually felt never paged anyone. Your SLO lives on a slide, and nothing changes when you blow through it. This skill turns SLOs into working reliability operations: burn-rate alerts that page only when the budget is really at risk, a budget policy with real triggers, an incident process people can follow at 3 a.m., and capacity numbers you can defend.

What it does

• Writes multiwindow, multi-burn-rate alert rules (PromQL or your stack's syntax) with page and ticket tiers, and converts budget percentages into minutes and bad requests.
• Drafts error budget policies with concrete thresholds, named decision owners, exceptions for dependency-caused burn, and exit criteria.
• Defines incident severities by user and budget impact, plus incident commander, comms, and ops roles, an update cadence, and postmortem triggers.
• Builds capacity models from peak load, per-instance throughput, utilization targets, N+1 headroom, and provisioning lead time.
• Catches broken premises: SLOs higher than their dependency chain allows, targets your on-call response time cannot meet, and low-traffic services where percentage alerts flap.
• Audits existing alerts and on-call load, then separates symptom pages from cause noise and ranks toil by hours saved.

What you get

Ship-ready artifacts sized to the request: an alert rules file, a one-page budget policy, a severity matrix and incident runbook skeleton, or a capacity table with the arithmetic shown and every assumption labeled.

Best for

Platform and SRE engineers, tech leads who carry the pager, and engineering managers setting up on-call or reliability policy for services that already run in production.