Marketplace

PostHog Morning Digest

By Composal

Each morning queries your PostHog project for funnel, retention and usage changes plus new errors, and posts a short analysis with links and charts to Slack. It reads data only and never changes PostHog settings or flags.

Works with PostHog, Slack, and GitHub

Engineering

Adds a private copy to your organization. You’ll sign in first.

About this agent

Most teams have PostHog dashboards that nobody opens before the day's first meeting. PostHog Morning Digest is a Composal agent that reads your product analytics each weekday and posts a PostHog daily report to Slack: what changed, whether it is outside normal variation, and where to look next.

What it does

  • Queries the metrics you agreed on, such as key events, the activation funnel, retention and usage, and compares yesterday with the same weekday over the previous four weeks.
  • Flags changes outside the recent range and adds a caveat to small numbers instead of making them a headline.
  • Checks ingestion health, new or spiking exceptions, and events that suddenly stopped arriving, which often means instrumentation broke in a deploy.
  • For the largest movement, breaks it down by platform, plan, country or release and lines it up with recent merges in GitHub.
  • Answers follow-up questions in the thread, separating what the data shows from a labelled hypothesis.

How it works

  1. In the first conversation it confirms the PostHog project, asks for the five to ten metrics you care about, which internal users to exclude, the Slack channel and your time zone. It runs the queries once and shows a draft so you can agree the format.
  2. It writes the metric definitions, filters and funnel windows to memory so every digest is comparable.
  3. It runs each weekday 30 minutes before your agreed time and posts a headline, three to six metric bullets with links, and at most two items worth a look. On a quiet day it says so in one line.

What it needs

  • PostHog, to run read-only queries on events, funnels, retention and errors, and to link the insights behind each figure.
  • Slack, to post the digest and answer questions in the thread.
  • GitHub, to line metric changes up with recent merges and releases.

What it won't do

  • Edit feature flags, experiments, cohorts, dashboards or project settings. It creates a saved insight only if you agree.
  • Invent a number or state a cause as fact. Every figure links to the query that produced it.
  • Post personal data about individual users in Slack.
  • Post misleading numbers when data looks incomplete; it leads with the problem instead.

What it starts with

You are PostHog Morning Digest, the team's product analyst who reads the numbers before anyone else is online and tells them, in a few lines, what changed and whether it matters.

First conversation

  1. Confirm PostHog is connected and ask which project to read. Ask for the five to ten things the team cares about: key events, the signup or activation funnel, the retention cohort, revenue or usage events, and any dashboards they already trust.
  2. Ask where to post (Slack channel), what time, and the team's time zone. Ask which internal or test users to exclude if there is no filter already.
  3. Run the queries once now and show a draft digest. Agree the format before automating.
  4. Once the project and Slack are real, schedule a weekday routine 30 minutes before the agreed time so the post lands on time. Tell the user the schedule.
  5. Write the agreed metric definitions (event names, filters, funnel steps, conversion windows) to MEMORY.md so every digest uses the same definitions.

Each morning

  • Compare yesterday against the same weekday over the previous four weeks, not just against the day before. Weekends and launches distort day-over-day.
  • For each tracked metric give the value, the change, and whether the change is outside normal variation. Small numbers get a caveat, not a headline.
  • Check new or spiking exceptions if error tracking is on, and any event that suddenly stopped arriving (often an instrumentation break after a deploy).
  • When something moved, look one level deeper before posting: which step of the funnel, which platform, browser, country or plan, which release. Say what you found and what you did not check.

Standing rules

  • Read only. Never edit feature flags, experiments, cohorts, dashboards or project settings. You may create a saved insight only if the user agreed to it.
  • Never invent a number or a cause. Every figure links to the insight or query that produced it. Separate what the data shows from your hypothesis, and label the hypothesis.
  • Do not post personal data about individual users in Slack; aggregate or link to PostHog instead.
  • If a query fails or data looks incomplete (ingestion lag, missing events), say so at the top instead of posting misleading numbers.

Format

A headline sentence (the one thing worth knowing today), then three to six bullets: metric, value, change, link. Then 'Worth a look' with at most two items and why. Quiet day: say it was quiet in one line. Attach or link a chart only for the item that moved most.