What problem does it solve?
Teams running PostHog logs face a firehose of millions of lines per hour where meaningful changes — a fresh error signature, a service going silent, a severity spike — drown in routine info-level noise. This Skill gives an AI agent a disciplined scouting workflow that surfaces only validated, window-over-window shifts and files them as actionable inbox reports instead of raw alerts.
Core Features & Use Cases
- Delta-based pattern detection: Anchors every run on
logs-patterns-diff to classify message templates as new, rate-shifted, or gone versus a baseline window, catching low-volume all-error signatures that share-based reads miss.
- Firehose-safe querying: Enforces severity- and service-filtered counts (
logs-count, logs-services-create, logs-count-ranges) so queries never time out on unfiltered streams, and avoids the -30m month-unit date footgun.
- Report authoring with dedupe: Checks the inbox and scratchpad memory before writing, edits existing reports for recurring issues, routes
suggested_reviewers to service owners, and records baselines under pattern:/noise:/report: keys for future runs.
- Use Case: A deploy introduces a new fatal-level template at low volume. The scout's diff read flags it as
new, corroborates it with bounded counts and an error-tracking cross-check, then authors a report with the service's chart series and the owning reviewer.
Quick Start
Ask the agent to run a logs scout over the last hour and report any new error patterns, volume bursts, or silent services it finds.
