signals-scout-web-analytics

Detects per-channel acquisition divergence, attribution breakage, and landing-page health steps in PostHog web traffic.

713|118|Updated Aug 11, 2020
One-click install
npx skills add https://github.com/PostHog/posthog-foss --skill signals-scout-web-analytics
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: signals-scout-web-analytics
Source: https://github.com/PostHog/posthog-foss/tree/main/products/signals/skills/signals-scout-web-analytics
Command: npx skills add https://github.com/PostHog/posthog-foss --skill signals-scout-web-analytics

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Aggregate traffic totals hide segment-level problems: one acquisition channel can collapse, UTM tagging can silently reclassify paid traffic into Direct, or a landing page's bounce rate can step up, all while the site's overall numbers look normal. This Skill continuously scouts PostHog web analytics data to surface exactly those segment-vs-aggregate divergences and authors actionable reports.

Core Features & Use Cases

  • Channel divergence detection: Scores each $channel_type against four seasonality-aligned windows (7/14/21/28 days back) using a median/MAD robust z-score, so a channel stepping away from its own baseline is caught while totals hold.
  • Attribution breakage detection: Identifies paired opposite moves (e.g., Paid Search down while Unknown/Direct rises) and falling tagged-session share, pointing to UTM stripping or referrer propagation failures.
  • Site-health monitoring: Watches per-host, ID-normalized entry paths for bounce-rate steps and traffic cliffs, plus 404/not-found event spikes with feeder-path attribution.
  • Report authoring and dedupe: Authors or edits inbox reports end-to-end via the report channel, with scratchpad memory (pattern:, noise:, dedupe:, report: keys) so recurring cadences and known issues are never re-filed.
  • Use Case: A marketing team's consent banner starts stripping UTM parameters. The scout notices Paid Search sessions dropping while Direct rises by a matching amount, confirms tagged share fell, and files a P2 report naming the cause, the fix, and a success criterion.

Quick Start

Ask the agent to run a web analytics scout pass over the last 24 hours of PostHog sessions and report any channel, landing page, or 404 diverging from its own baseline.

Frequently Asked Questions about signals-scout-web-analytics

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I detect a traffic drop in one acquisition channel in PostHog?

Query the sessions table grouped by $channel_type and compare the last 24 hours against the same 24-hour windows 7, 14, 21, and 28 days back. Score the divergence with a robust z-score using the windows' median and MAD so normal weekly rhythm does not trigger false alarms.

How to tell if UTM tagging broke instead of losing paid traffic?

Look for paired opposite moves: a paid or campaign channel dropping while Direct or Unknown rises by a similar amount with the same onset. Confirm by checking whether the share of sessions with any $entry_utm_source set fell across aligned windows while total traffic held.

Why compare web traffic to 7 days ago instead of yesterday?

Web traffic is strongly day-of-week seasonal, with weekdays often running two to three times weekend volume. Comparing the same 24-hour window 7, 14, 21, and 28 days back aligns both weekday and time-of-day, so a real step diverges from every aligned window.

Does PostHog have a built-in 404 event for broken link monitoring?

No, PostHog has no native 404 event; teams instrument their own. Discover the project's convention by searching recent events matching patterns like '%404%' or '%not%found%', then baseline its daily volume and watch for spikes with feeder path and referrer attribution.

When should a traffic surge not be reported as growth?

A surge needs provenance checks first: real referred sessions show plausible session durations, pageview counts, person spread, and library mix. Hundreds of zero-duration single-pageview bounces from one unfamiliar domain indicate referrer spam, which should be recorded as noise, not celebrated.