What problem does it solve?
Setting up log alerts by guessing thresholds leads to noisy alerts that fire constantly or silent alerts that never fire. This Skill turns alert authoring into a measurement problem: it triages services, characterizes their 7-day baseline, drafts thresholds from real percentiles, and back-tests each alert against history before shipping it.
Core Features & Use Cases
- Service triage: Ranks the top 25 services by log volume and error rate to identify which ones are worth alerting on.
- Baseline-driven thresholds: Pipes count-range data into a statistics script that computes p50/p95/p99 and suggests a threshold scaled to the alert window, with health flags for sparse, flat, or spiky baselines.
- Simulation before shipping: Replays each draft alert against 7 days of history with a full state machine, iterating up to 3 rounds to land fire counts in the 0-3 per week band.
- Use Case: A user asks to "set up alerts for my project." The Skill triages services, finds api-gateway has a spiky error baseline, drafts a 5-minute 2-of-3 alert at the suggested threshold, simulates it firing twice over 7 days, then creates the alert wired to a confirmed Slack channel.
Quick Start
Set up low-noise error alerts for the services in my PostHog project and suggest which ones are worth monitoring.
