What problem does it solve? During a live outage, engineers often jump to conclusions based on gut feel or a single log line, leading to wrong fixes and prolonged downtime. This Skill enforces disciplined incident triage by collecting evidence in parallel, scoring hypotheses against a visible scorecard, and requiring proof before declaring a root cause. ## Core Features & Use Cases - Parallel Evidence Collection: Launches five concurrent tracks (recent changes, service health, topology, telemetry, runbooks) across AWS, Kubernetes, Azure, New Relic, and Confluence. - Hypothesis Scorecard: Normalizes every finding into typed signals (VERIFIED, CORRELATED, SUSPECTED, VERIFIED_NEGATIVE), scores the top 5 hypotheses, and automatically eliminates contradicted ones. - Root Cause vs Mitigation Separation: Enforces hard rules so irreversible mitigation always requires a confirmed root cause, with a SEV1 fast path for reversible mitigation. - Use Case: When push notifications stop working across multiple school products, run the skill with the product and symptom; it correlates CloudTrail changes, ECS health, and telemetry to pinpoint a shared dependency failure and produce a remediation plan. ## Quick Start Ask the assistant to triage an incident by running /ai-sre:sre-incident with a product name and symptom, for example "psw-us push notifications failing since 14:00 UTC".
