When something looks wrong

The short list of symptoms, what each usually means, and where to look before you worry.

You will be able to match a symptom to its usual cause and know where to look before you worry.

Most problems on a mineral present as one of the symptoms below, and most have a one-line cause. This page is the list, ordered by how often each actually happens rather than by how alarming it feels.

The morning brief did not arrive#

Look at Connections first. A scheduled job that needed a dead connection fails rather than pretending, so a missing brief usually means an expired sign-in, and the row will say so.

If Connections is green, check the schedule itself on Skills: a malformed schedule is refused rather than guessed at, and the page says so.

Third check, if both of those look right: the fold at the bottom of Connections for services connected to your Claude account. Those work in your chats and can never run in a scheduled job, so a brief that depends on one keeps failing until the service is connected on the mineral itself, from the Connections page.

My assistant answers like it knows nothing about me#

It probably does not. Check the Onboarding card on Overview: until the interview is finished, every answer is generic. That is the cause almost every time this is reported.

I connected the Claude Code app but the checklist still says not signed in#

That is the expected result, not a failure. Connecting the app signs in your laptop and reaches into your mineral over SSH; it does not sign the mineral in, so it does not switch on scheduled jobs and nothing it does makes the Overview checklist tick. The app now says it in bold: "This is an extra, not a step." The checklist item belongs to the Terminal tab: open it and run claude once.

My assistant uses a service when I chat, but the scheduled job cannot see it#

The service is connected in a place only chats load. A connection made inside a Claude Code session works when you talk and is invisible to scheduled jobs, and Connections says so rather than hiding it: the row is marked "in your chats only, not in scheduled jobs" and carries a one-click "Use in jobs too" button, which is the fix. The one kind with no button is a service connected to your Claude account in the app's own fold: those can never run in jobs, because there is nothing on the mineral to move; connect the service on the mineral instead.

The app says "software unknown" for my mineral#

The mineral is running a build too old to report which build it is. The fix is an update: open Help, and under Mineral software run Update and restart this mineral. When it comes back it will be current and say so; how updates reach you is the mechanics.

The mineral's software is behind the app. The Connections page fills in the addresses it signs in against from what the mineral reports, and a mineral running older software does not report them, so the sign-in starts with a blank address and fails with a raw error instead of a kind sentence. The fix is the same update as above: open Help and run the software update under Mineral software. Once the mineral is current, Sign in behaves.

My new computer cannot connect#

Adding a computer needs a machine that already opens the mineral to vouch for it: add another computer walks both halves, and its last section covers the refusals (a name that does not match the bundle, a mineral that did not answer the probe). If the mineral is mid-restart, the bundle stays good; try again in a minute.

Google shows a scary "unverified app" screen#

That screen is correct and expected: the app is yours, you created it minutes ago, and nobody was asked to verify it. Click Advanced, then Go to your app's name. The whole story is in connect Google.

Rare, and real: Google dropped the key your mineral holds, so "your scheduled jobs cannot reach Gmail or Calendar until you sign in again". The mineral checks the key itself and sends that message exactly once, not on repeat. The fix is the two clicks the message names: Connections, then Google, then Sign in. If the sign-in itself misbehaves, connect Google has the whole story.

There is a machine on my roster I do not recognise#

Remove it, then work out what it was. A mineral made through the setup wizard is born with only your own key, so every row should be a machine of yours, a machine you added, or a live support grant you made. See machines that can open your mineral.

A machine appeared on my roster that I was already using#

If the notice reads "This computer was already connected but had never been listed here, so it has been added to the roster", that is the roster healing itself, not an intruder: a machine connected before the roster existed gets listed the first time you open the page on it, and the new row is yours to rename or revoke like any other. A machine you genuinely do not recognise is the entry above, and gets the opposite treatment.

Secrets lists things I never added#

Rows described as the mineral's own plumbing are exactly that: the keys and tokens the mineral needs to be a mineral (status keys, internal tokens, and so on). They are listed because a Secrets page that hid some of the secrets would be a different kind of page. There is nothing on them for you to revoke; each row that IS yours says where it comes off.

My bot never replies on Telegram#

Look at the Telegram card on Connections. A chip reading "one step left" means the pairing never finished: the bot exists, but it was never bound to your chat, so it has nobody to answer. The pairing code expires after ten minutes, so reopen the card for a fresh one and message that code to your bot. The button you want is Finish setup, not disconnect; connect Telegram walks the two steps.

Everything disconnected for a minute#

That is the expected face of a software update. The restart takes about a minute and everyone connected drops briefly: this panel, open terminals and Claude Code sessions all reconnect on their own afterwards, and nothing of yours is touched.

The mineral is unreachable#

Sometimes it is genuinely the machine, and sometimes it is between you and it. Give it a minute, then open Help and run Update and restart this mineral (under Mineral software), which restarts the mineral's software without touching anything of yours. If it stays unreachable, the machine itself is worth a look in your own hosting console, Hetzner's or DigitalOcean's (it is your server; their console can reboot it), and a person helping you can go deeper with your permission.

Something else#

Report it, and say what you saw rather than what you concluded: "the brief had no calendar section" finds the cause faster than "the assistant is broken". When a fix lands, the app's version notes say so in plain words.