Skip to content

Verify your setup

Every connection in Data Workers is in exactly one of three states, and the state is earned, not assumed:

StateWhat it meansWhat you do
🟡 Evaluation (sample data)The agent works out of the box on built-in sample data so you can try it before connecting anything.Nothing — this is the default.
🟢 ConnectedThe agent is wired to your real system and a live test has passed to confirm it works.Provide credentials, then verify.
🔴 Needs attentionIt was connected, but the connection is now failing — expired token, changed permissions, network.Re-check the credential.

Setting a credential is not the same as a working connection. A system with credentials configured but no passing test stays out of the 🟢 state on every surface — terminal and console alike. This verify-before-trust step is deliberate: it’s how you (and we) can honestly say an agent is running on production data rather than assuming it because a variable was set.

From your coding agent — ask for a live test, per system:

Test the connection to my Snowflake catalog.

The agent makes a real call with your credential and reports 🟢 or 🔴 with the failing step named. To see everything at once:

What’s the status of my data connectors?

From the Spellbook console (Scale) — press Test connection on the connector card. Same rule: the card only turns green on a passing live check.

🔴 always names a cause — the most common are an expired token, a permission that was tightened, or a network path that changed. Fix the credential or grant, then re-run the test. The Troubleshooting page maps the frequent failure messages to fixes, and each connector page lists the minimum grants the test needs.

Sample-data mode isn’t a limitation to apologize for — it’s how you evaluate a fleet of agents without handing credentials to a vendor on day one. And the hard line between “credential set” and “verified live” is how a platform that writes to production systems stays trustworthy: if the state says 🟢, a real test passed, full stop.