How verification works

Four trust levels, applied per entry and shown with a date. Claims are labeled Official or Independent based on where the evidence comes from.

Trust levels

Official

Demonstrated or published by xAI / Cursor sources; linked to the official page.

Tested

We ran the setup ourselves and record what happened, with evidence.

Community

Shared publicly by its creator; attributed and linked, not re-run by us.

Untested

Listed with provenance but no run evidence. Never displayed as blank — always labeled.

Risk badges

Assigned by analyzing the bot's published prompt text (never by running it with your data). A badge means the capability is present, not that the bot is malicious. Missing approval gates are flagged separately.

sends

The prompt instructs the bot to send email/messages on your behalf.

spends

The bot can spend money or trigger refunds/purchases.

logins

Setup requires handing the bot credentials or connected accounts.

browses

The bot operates a logged-in browser session.

Last updated 2026-08-31. Changes to this methodology are logged in /corrections.