Every creator has that moment: you download the CSV from your dashboard, open it, and stare at forty rows of numbers. Which posts worked? Is the channel actually improving, or did one lucky video skew everything?

I built a Hermes skill, creator-analytics-digest, for exactly this. You give your agent the exported file. A small script does the counting, because arithmetic is where chatty assistants make mistakes. It finds the title, date and metric columns on its own (English, French and Spanish headers work), ignores "Total" rows, and understands values like 1,200, 4.1% or 2.4K.

Then the agent writes the digest: what worked, what lagged, the trend, and three follow-up ideas, each tied to a real row. The skill tells it to label reasons as guesses, since an export cannot tell you why people clicked.

A short example from a fictional knitting channel:

  • Median views: 1,500
  • Standout: "Sweater pattern", 3,500 (more than twice the median)
  • Median of the second half of the period: 91% above the first half

The suggestion: a follow-up with another concrete pattern, flagged as a hypothesis to test, not a rule.

Nothing leaves your machine: no API key, no upload, only Python's standard library. Limits are stated in the skill: a few dozen rows give noisy results, and I only tested it on small fictional files, not on every platform's real export.

Try it on your last export.

Skill (MIT): https://github.com/RobinBeraud/hermes-skills/tree/main/data-analysis/creator-analytics-digest
All skills: https://github.com/RobinBeraud/hermes-skills

This post was written with AI assistance and reviewed through automated checks.