When something does not work

The failures that have real causes, what each one actually means, and what to check first.

Most of what looks broken here is one of a dozen known situations. This page is those, with what each one means.

The panel is not there at all

The extension only runs on *.service-now.com. An instance on a customer's own domain is outside what the browser granted it, and nothing will appear.

If the address is right, the page may have loaded before the extension woke up. Reload once. The widget is injected on request rather than declared, so a worker that was asleep can miss a page that was already loading.

The palette does not open with /

You are in a field. That key is vetoed inside inputs, text areas, selects, anything editable, and inside the platform's code editors, because there / is text. Click outside the field and press it again. / is the only key that opens the palette.

Alt+Shift+N or Alt+Shift+D does nothing

Your browser did not assign the key to the extension. A suggested key is only a suggestion, and on a fresh Chromium profile Alt+Shift+N was not assigned. Set it in your browser's extension shortcuts page, or use /tn and /tt. Details are on the keyboard page.

A command says the instance refused it

A 401 or 403 is your ServiceNow account, not the extension: it asks with your session and reports what it is told.

If you are impersonating, the message says so, because that is the most confusing version of this failure. The request ran with the other person's rights. /unimp stops that.

A list came back with more rows than I asked for

That is the failure switches exist to prevent. ServiceNow silently drops a condition it does not understand and runs the query anyway, so an unknown switch, or a field the table does not have, is refused here rather than passed on. If you see it happen anyway, the condition came from somewhere else.

Dark mode is half applied

Three generations of interface are being repainted, and the legacy pages and the Service Portal do not read design tokens at all, so they are handled by hand. A page that arrived after we last looked can have a corner that was missed. That is a gap in a list rather than a broken mechanism, and it is worth reporting with the page's address.

One class of gap is structural rather than missed: a decoration drawn behind the page's own background disappears once that background is opaque. Flow Designer's connector lines were exactly that.

A background script ran and did nothing

HTTP 200, no "Script completed", no records. That is the signature of a run whose submit field never reached the platform. From our panel that is a bug and worth reporting; the same symptom on the platform's own page usually means the script threw before it printed anything.

If instead you were signed out, that is the CSRF token being rotated by a second run. Only one run should be in flight at a time, and the panel enforces that across tabs.

The assistant answered slowly, then failed

Its budget is 165 seconds in the browser against a 150-second ceiling on the server, so the server's own message usually wins the race and tells you what happened. A timeout means the work may still be running; asking again is safe, and asking about less of the record at once is faster.

The assistant is temporarily unavailable is the model provider failing. The real reason is logged with secrets removed and deliberately not shown, because provider errors quote credentials back at you.

The vault says it cannot attach a credential

Three different things, and the message says which. Locked: unlock it in settings. No credential: there is no entry for that host. Host mismatch: the entry's host and the request's host are not the same, which is the check that stops one customer's credential being sent to another's instance.

An alignment target cannot be reached

Only *.service-now.com hosts can be checked, for the same permission reason as above. On an SSO-only instance choose the browser session rather than a stored credential: there is no ServiceNow account to store.

A recorded test reports a warning rather than a pass

A warning means the value went in and the field changed it, for example a date rewritten into the instance's format. Broken means the value was ignored. An element found only by its position in the page is also a warning, because that run will not survive a redesign.

Nothing is rolled back: the records a replay created stay, and the report lists them.

Nothing is being captured

Check that Auto-Insert is on and not paused. Then check that the active update set is not Default, and not already complete: the Default-set guard should be saying so. Auto-Insert reacts to saves, so a record changed before you switched it on is not captured. Orphans finds those.

A personal rule did not fire

A rule fires only for the background kind, only on the exact table its trigger names (a child table does not count), and only while Personal rules is on in the panel's Settings tab. A condition that throws skips the rule and says so in the Logs tab.

My scripts or wildcards are gone

They belong to a profile, and switching profile switches the whole library. Check the profile shown at the top of the panel.

A wildcard will not expand

The expansion happens when you type the closing bracket. Pasting [trigger] does not count.

Pro says it is in use on another device

Pro holds two devices at a time. Use it here takes it for this one, and the device that used it least recently loses it. Pro is on three devices already means a fourth was refused: remove one under Account on the site. Both are explained in plans.

Saving says I cannot add more

That is the plan's limit on profiles, saved scripts or wildcards, and the message names which one. What you already have is kept.