Auto-document your captured event taxonomy

Why this matters

Event taxonomies always drift. The team agrees on signup_completed in week 1. Someone on the mobile team ships signup_complete (no d) in month 3. The web team renames to user_signed_up in month 6. By month 12 you have three names for the same conceptual event, four schemas of properties, and the only person who knows which one to query has left.

The fix: a living event-taxonomy document. The trouble: nobody writes one because it goes stale by next quarter. The fix to the fix: auto-generate it from the actual event stream.

Agentry has agentry_list_event_names — every event captured plus its count, first-seen, last-seen, and sample properties. One prompt produces a clean Markdown table that pastes straight into Notion or Linear. Re-run weekly, the doc is always current.

What you get

• A Markdown table of every event name with count, first-seen, last-seen, and sample properties
• Sorted by volume (so you see the heavy hitters first)
• A “deprecation candidates” section flagging events with zero activity in the last 7 days
• Quick spot for typos (signup_completed vs signup_complete)

Walk through it

agentry_list_event_names { project_id: "default" }

{
  "server_emitted": [
    { "name": "case.created" }, { "name": "case.resolved" }, { "name": "deploy.recorded" }
  ],
  "analytics_events": [
    { "name": "page_view",          "count": 1240000, "first_seen": "2024-08-12", "last_seen": "2026-05-16" },
    { "name": "signup_completed",   "count":    8400, "first_seen": "2024-08-15", "last_seen": "2026-05-16" },
    { "name": "checkout_started",   "count":   42000, "first_seen": "2024-08-20", "last_seen": "2026-05-16" },
    { "name": "order_completed",    "count":   18400, "first_seen": "2024-08-20", "last_seen": "2026-05-16" },
    { "name": "product_viewed",     "count":  410000, "first_seen": "2024-08-20", "last_seen": "2026-05-16" },
    { "name": "signup_complete",    "count":      62, "first_seen": "2025-12-01", "last_seen": "2026-05-15" },
    { "name": "ab_assigned",        "count":  118000, "first_seen": "2025-04-12", "last_seen": "2026-05-16" },
    { "name": "video_uploaded_v2",  "count":     180, "first_seen": "2024-11-01", "last_seen": "2025-09-12" }
  ]
}

agentry_analytics_query {
  project_id: "default",
  query: "
    select
      event,
      arrayDistinct(arrayFlatten(groupArray(JSONExtractKeys(properties)))) as sample_keys
    from events
    where timestamp > now() - interval 30 day
    group by event
  "
}

{
  "rows": [
    { "event": "page_view",        "sample_keys": ["path", "referrer", "user_agent"] },
    { "event": "signup_completed", "sample_keys": ["source", "plan", "country"] },
    { "event": "checkout_started", "sample_keys": ["cart_value", "item_count", "shipping_method"] },
    { "event": "order_completed",  "sample_keys": ["order_id", "total", "items", "coupon_code"] },
    { "event": "product_viewed",   "sample_keys": ["sku", "category", "price"] },
    { "event": "signup_complete",  "sample_keys": ["source", "plan"] },
    { "event": "ab_assigned",      "sample_keys": ["flag_key", "variant"] },
    { "event": "video_uploaded_v2","sample_keys": ["video_id", "size_bytes", "format"] }
  ]
}

# Event taxonomy (auto-generated 2026-05-16)

| Event | 30d count | First seen | Last seen | Sample properties |
|---|---:|---|---|---|
| `page_view` | 1,240,000 | 2024-08-12 | 2026-05-16 | path, referrer, user_agent |
| `product_viewed` | 410,000 | 2024-08-20 | 2026-05-16 | sku, category, price |
| `ab_assigned` | 118,000 | 2025-04-12 | 2026-05-16 | flag_key, variant |
| `checkout_started` | 42,000 | 2024-08-20 | 2026-05-16 | cart_value, item_count, shipping_method |
| `order_completed` | 18,400 | 2024-08-20 | 2026-05-16 | order_id, total, items, coupon_code |
| `signup_completed` | 8,400 | 2024-08-15 | 2026-05-16 | source, plan, country |
| `signup_complete` ⚠ | 62 | 2025-12-01 | 2026-05-15 | source, plan |

## Deprecation candidates (0 activity in last 7 days)

- `video_uploaded_v2` — last seen 2025-09-12 (180 lifetime events)

## Typo / duplicate candidates

- `signup_complete` (62 events) vs `signup_completed` (8,400 events)
  → recommended fix: drop the `signup_complete` callsite, all callers should be on `signup_completed`

Want me to set up a Routine that regenerates this every Monday and posts it to a Notion page?

The output

Event taxonomy auto-doc — 2026-05-16

Top events (by 30d volume):
  page_view              1,240,000     props: path, referrer, user_agent
  product_viewed           410,000     props: sku, category, price
  ab_assigned              118,000     props: flag_key, variant
  checkout_started          42,000     props: cart_value, item_count, shipping_method
  order_completed           18,400     props: order_id, total, items, coupon_code
  signup_completed           8,400     props: source, plan, country
  signup_complete ⚠             62     props: source, plan         ← TYPO

Deprecation candidates (no activity in 7d):
  video_uploaded_v2   last: 2025-09-12   (180 lifetime)

Typo / duplicate candidates:
  signup_complete (62)    vs signup_completed (8,400)
    → kill the misnamed callsite

Output saved as Markdown — paste into:
  Notion: agentry/events
  Linear doc: agentry-event-taxonomy
  Slack: #data

Setting it up

No setup at all — this is a pure read against the events your app is already sending. If you’re getting “no events found,” it just means analytics isn’t flowing yet. Send a few test events to confirm:

await fetch(`https://api.agentry.sh/v1/analytics/${PROJECT_ID}/`, {
  method: "POST",
  headers: {
    "Authorization": `Bearer ${process.env.AGENTRY_DSN}`,
    "Content-Type": "application/json",
    "User-Agent": "myapp/1.0",  // REQUIRED — Cloudflare 403s default UAs
  },
  body: JSON.stringify({
    event: "signup_completed",
    distinct_id: "[email protected]",
    properties: { source: "homepage", plan: "free" },
  }),
});

Once events flow, re-run the prompt. If you want this to be a permanent artifact, two options:

Option A: Weekly Routine that posts to Slack. Same pattern as weekly-agent-digest, regenerates the table every Monday.

Option B: Embed as a public dashboard. Publish via agentry_publish_query and iframe-embed in your Notion/wiki — that way it’s always current without re-paste.

Variations

• “Same table but include sample event values (top 5 values per property), not just keys. Helps spot bad data.”
• “Just the typo / duplicate candidates — I want a clean cleanup list.”
• “For each event, show me how many unique distinct_ids fire it — events with low cardinality are often background-job spam.”
• “Weekly Routine: regenerate, diff vs last week, post deltas to #data (new events created, events that died, count changes).”