Free agent skill

Get the Semantic View Canary skill

Works in Claude Code or Codex, on any Snowflake edition.

For skill updates: we’ll add you to a read‑only GitHub team once it opens.

Optional follow-ups

Both choices are optional. You get the link with neither selected.

Keep a Snowflake Semantic View Accurate: Review It, Then Build a Canary

Use Claude Code to check what an AI-built Snowflake semantic view really computes, then build a native canary that emails you when a validated number moves.

Who it's for: Data and analytics engineers running Snowflake semantic views, or about to let AI build one.

The problem: A view can map revenue to gross when people mean net, run clean, and answer about 5% high. Data also changes underneath a correct view.

Takeaways:

  • Review the DDL four ways: keys, business terms, term to SQL match and comments.
  • Cortex Agent evaluations catch a moved number only when the saved answer is tighter than the move.
  • A Task, a log table and an Alert email you when a validated number moves.

Resources

What you get

The skill reads your semantic view, asks for the numbers you’ve already validated, and writes the Task, log table and Alert SQL for you to review and run. Every day the canary asks the view for those numbers on a closed period, keeps each answer, and emails you when a number moves or a check never lands.

Installing it

The quick way is to let your agent do it. Open Claude Code or Codex and ask, using the path where your download landed:

Install the agent skill from ~/Downloads/semantic-view-canary-skill.zip into my skills folder

To do it by hand, unzip the download into ~/.claude/skills/ for Claude Code or ~/.codex/skills/ for Codex. Then ask your agent to set up a canary for your semantic view.