Write down in a few lines what you just tried, what you expected and what
happened instead. Name the file or the page as well.
Sometimes the assistant gets stuck. A connection that will not work, an error that keeps coming back, or a feature that behaves differently than you expected. That is when you want a human. The ⋯ menu of the assistant has Ask Invoker for help for exactly that.
It is not a contact form that drops into a bucket on our side. The message arrives with your account, your link status and the last log lines of your own account attached, so we do not need three emails first to work out where to look.
Three types of message
Above the text field you pick the type. That decides how we pick it up.
| Type | When | What we do with it |
|---|---|---|
| Need help | You want to get something done and cannot | We reply with the steps, or do it for you |
| Something does not work | The behaviour is not what it should be | We try to reproduce it on our own setup |
| Idea or improvement | You are missing something in the assistant | It goes on the list; many features started this way |
You also fill in a subject and your message. The subject can be 140 characters, the message a little over 8000. So there is no need to be brief.
What is attached automatically
Along with your message we add, without you having to do anything:
- Your account and the server it lives on.
- The website you had open in the assistant at that moment.
- Where the message came from: the assistant, the panel or the terminal.
- The version of the plugin running on your server.
- The link status of Claude, Codex and GitHub, with the email address you are connected with.
- The last log lines of your account. Twelve at most, and only lines from your own account.
That last one is the most useful in practice. Nine times out of ten it already says where things went wrong, and if it does not, we know immediately that we have to look somewhere else.
You get a copy by email at the address on your account. If we reply, that reply lands in the same thread, because your address is also set as the reply address on the message.
What you should add yourself
Attaching things automatically does not replace what only you know. Three things make the difference between a quick answer and a back and forth:
- What you were trying to do, in plain words. Not "it does not work" but "I wanted to publish the preview to the live site".
- What you expected and what happened. Quote the error message literally if you saw one, even if it makes no sense to you.
- Whether it worked before, and when it last went well.
Feel free to copy in an instruction or a message from the assistant. The more literal, the better.
What you leave out
No passwords. Not for your account, not for WordPress, not for a database. It says so above the form, and it goes for keys and tokens too. We do not need them: we can look on our own servers without your password.
If the assistant shows you a password, after an installation for instance, mention in your message that you have it, not what it is.
Five messages per hour
You can send at most five messages per hour. At that limit you get the message "Please wait: at most five messages per hour". The limit is not there to slow you down but to stop a script or a stuck page from pushing hundreds of emails out the door.
Five is plenty in practice. Rather send one message with everything in it than five separate ones with a line each.
Common mistakes
"I only send a subject." Technically allowed, but it almost always earns you a question back. Write out the three things from the section above and we are a round quicker.
"I cannot attach my screenshot, so I skip it." True for this form, but you can email the screenshot separately as a reply to the copy you receive yourself. That copy is already attached to the right thread.
"I filed it under the wrong type." Not worth thinking about. Pick whatever comes closest; we read the message anyway.
"I will include my password so they can log in right away." Do not. Change the password if you already sent it.
"I will wait until I have figured it out myself." A report saying "I do not know where to look" is a perfectly good report. The log lines come along regardless, and that is where our search starts anyway.
If you would rather keep looking yourself first, the assistant stalls: resume, undo and retry covers what to do when an instruction hangs halfway.