── SKILL
event-taxonomy-builder✓ APPROVED

Name the moments that signal value in your product.

Interviews you about your product, names the events that mean real value, tiers each by reliability, and rolls them into a short capability list your analyst can score.

In plain English: “What should my product analyst actually measure?

Inside: The value-moment finder, the Tier 1/2/3 reliability test, the capability rollup, and the adoption rubric.

01 / HOW TO USE

Install it in one line, or paste it in.

1
In Claude Code (one command)
Copy the install line, paste it into your terminal, and restart Claude Code. The skill installs itself to ~/.claude/skills/ and runs automatically when it is relevant.
2
In Claude, ChatGPT, or a Project (no terminal)
Open the file, then upload it to your chat or paste its contents in. A skill is just a markdown file of instructions, so any capable AI can follow it.
New to skills? A skill is a plain-text file that teaches your AI a workflow. Point any capable assistant at it and it follows the steps, on your data.
02 / MAKE IT YOURS

Connect your context. Set it to your motion.

CONNECT YOUR CONTEXT · AND WHY IT HELPS
a product-analytics tool

it reads your live event stream, so it names real events instead of ones you hope exist, and flags the ones that barely fire.

a data warehouse

it checks event volume and reliability across the whole base before it trusts an event.

your onboarding or activation docs

it maps the taxonomy to the journey you already designed.

SET IT TO YOUR MOTION

This was built to be product-agnostic. Set these to your product:

The taxonomy is yours. The skill's job is to help you find the 5 to 15 moments that actually predict retention and expansion in your product, and to throw out the ones that just look busy.

Set thisWhat it isDefault / Example
ANALYTICS toolyour product-analytics connectorAmplitudeMixpanelPostHoga warehousea CSV
VALUE_MOMENTSthe actions that signal real valueyour own list (the skill helps you find them)
CAPABILITIEShow events roll up into features5 to 15 named capabilities
RELIABILITY_BARvolume/consistency an event needs to be trustedfires for most active accountsmost weeks
GRAINhow you group users into an accountdomainworkspace idorg id
ANTI_SIGNALSevents that look like value but are notloginspage viewspassive opens
03 / THE FULL SKILL

Everything the skill does, in full.

── WHAT THIS DOES

Turns "what should we actually measure?" into a taxonomy your team can trust. It interviews you about your product, names the events that signal real value (not vanity clicks), tiers each one by how reliably it means what you think it means, and rolls the whole set into a short list of capabilities an analyst can score. The output is the source of truth every downstream product read runs on, so two people asking "how are they using us?" get the same answer.

── THE METHOD
  1. 1
    Name the value moments

    Start from the outcome, not the feature. Ask what a customer has to do repeatedly for the product to be working for them. Those repeated, outcome-linked actions are your candidate value moments. A click is not a value moment. A completed core action that ties to why they bought is.

  2. 2
    Tier every event by reliability

    Every candidate event gets a tier: TIER 1 (fires cleanly, means exactly one thing, high volume), TIER 2 (useful but noisy or lower volume, use with care), TIER 3 (a trap, looks like value but is not, or double-counts a Tier 1). Name the traps out loud so no one builds a read on them later.

  3. 3
    Roll events into capabilities

    Group the trusted events into a short list of capabilities, the features that actually signal value. Each capability maps to its raw events explicitly, so an analyst can score adoption per capability instead of drowning in event names.

  4. 4
    Set the adoption rubric

    Define what POWER, ESTABLISHED, EMERGING, DORMANT, UNTOUCHED, and NEVER-ADOPTED mean for each capability, in your product's real numbers. This is the rubric every product read reuses, so adoption means the same thing every time.

  5. 5
    Flag the anti-signals

    Call out the events that inflate a picture: logins, page views, passive opens, anything that rises when nothing of value happened. A baseline "any activity" signal is fine as a floor, never as proof of value.

── QUALITY GATES
  • Every trusted event has a named tier and a one-line reason. No untiered events in the taxonomy.
  • Traps are named explicitly, not quietly dropped.
  • Each capability lists the exact raw events it rolls up, never a vague label.
  • The adoption rubric is in your product's real numbers, not borrowed defaults.
  • No event is trusted on volume it does not have. A rare event is Tier 2 or 3, never Tier 1.
── OUTPUT (EXAMPLE)
EXAMPLE OUTPUT
EVENT TAXONOMY · [your product] · built [date]

Capability          Raw events (trusted)              Tier
Core action         action_completed                  1
Automation          rule_created, rule_ran            1
Templates           template_used                     2 (noisy: counts drafts)
AI assist           ai_compose_accepted               1
Integrations        integration_connected             2 (low volume)

TRAPS (do not build reads on these):
  - session_start   looks active, means nothing about value
  - page_view       inflates any "usage" number

Adoption rubric (Core action):
  Power >= 5 completions/week · Established 1-4 · Emerging <1 ·
  Dormant used then stopped 14d+ · Untouched never fired

Next: instrument ai_compose_accepted server-side; it is Tier 1 logic
but currently client-only, so ~15% of fires are lost.
── EXAMPLE CHAT
EXAMPLE CHAT
You
We're a scheduling tool. People connect a calendar, create booking pages, and get meetings booked. What should my product analyst measure?
CLAUDE
Start from the outcome, which is a meeting actually booked, and work backward. Here's the first cut: Capability Trusted event Tier Why Activation calendar_connected 1 one-time, but the gate to all value Core action booking_page_created 1 the thing they came to do Value realized meeting_booked 1 the outcome; this is your north star Sharing page_link_shared 2 predicts bookings but noisy The one I'd flag: do not measure `login` or `dashboard_view` as usage. A customer who logs in daily but never gets a meeting booked is not healthy, they just visit. That is the ghost-active trap, and building a read on logins is how teams miss churn.
You
We also have team plans. Does that change it?
CLAUDE
It adds a grain question. Roll events to the workspace, not the person, so a team with one power booker and five dormant seats reads as partial adoption, not full. I'd add a `seats_active_ratio` capability at Tier 2. One sharpener: connect your analytics tool and I'll check real fire volumes, so I can confirm `meeting_booked` is clean and not double-firing on reschedules before you make it your north star.
── WHERE THE NUMBERS COME FROM

The tiers, the reliability bar, and the adoption thresholds are yours to set. The examples here suited a weekly-rhythm SaaS product. If your product is used in bursts or on a monthly cycle, retune the thresholds. The logic, name the outcome, trust events by reliability, roll up to capabilities, does not change.

04 / GO FURTHER

Where an operator takes this next.

The taxonomy is step one. Here is where an operator takes it once it holds.

You define value once, and every downstream read inherits it.

01
Wire it into a live product read

Feed the taxonomy to a product-usage analyst skill so every account gets scored on your real capabilities, on demand or nightly from Amplitude or your warehouse.

02
Watch reliability drift

Have a scheduled Claude task re-check event volumes weekly and flag when a Tier 1 event starts dropping fires, so instrumentation bugs surface before they corrupt every read.

03
Publish it as the shared source of truth

Push the taxonomy to Notion or your docs so product, CS, and sales all score adoption the same way instead of arguing about definitions.

05 / PART OF A BIGGER PLAY

One skill is the on-ramp.

── GET NEW SKILLS IN YOUR INBOX

One operator-built skill a week. Free.