What problem does it solve?
Monitoring distributed tracing data manually across many services makes it easy to miss real regressions or to drown in false positives caused by traffic spikes. This Skill watches per-service RED metrics (rate, errors, duration) in PostHog APM span data and files a report only when a validated regression clears the bar.
Core Features & Use Cases
- Regression Detection: Compares per-(service, operation) error rates and p95 latency against a seasonality-matched baseline from 7 days ago, guarding against volume-driven false positives.
- Root-Cause Investigation: Pulls representative failing traces, reads exception attributes, breaks down error signatures by attributes like service.version, and localizes latency via span trees and duration histograms.
- Report Authoring & Memory: Authors or edits inbox reports with evidence and charts, and persists baselines, noise patterns, and dedupe pointers in a scratchpad so future runs stay cheap and non-duplicative.
- Use Case: A payments service's p95 latency jumps from 320ms to 1.4s while request volume holds steady; the scout confirms the onset bucket, bundles the affected operations into one report, and routes it to the service owner.
Quick Start
Ask the agent to run the APM signals scout against your PostHog project to check for error-rate or latency regressions across your instrumented services.
