Your AI assistant can run your help desk — here's the free skill that teaches it how
If you use Claude, you can connect it to a ticketing workspace and let it file, sort, update and close tickets for you — by describing what happened in a sentence, the way you’d tell a colleague. This page is about the document that makes it good at that, rather than merely capable of it. It’s free to download and takes about ten minutes to set up.
First: what is a “skill”?
A skill is not software. Nothing installs, nothing runs, nothing can crash. It is a written document that your AI assistant reads — the same way a new hire reads the staff handbook on their first morning.
That distinction matters, because it explains what the skill is for. Your assistant is already smart enough to use a ticketing system. What it doesn’t know is your system’s house rules: which fields are compulsory, what the categories actually mean here, which things it must never do without asking. None of that can be worked out from the software alone — someone has to write it down.
That’s the skill. Written down, once, so you never have to explain it again.
What goes wrong without it
An assistant with access but no handbook is confidently, quietly wrong. Real examples, all of which we watched happen:
- It pastes the entire problem description into the ticket’s summary field. Now every list, search result and report shows a wall of text where a one-line title should be — permanently.
- It tries to mark a ticket “high priority” directly. Priority here is calculated from how urgent and how widespread the problem is, so the setting silently doesn’t stick.
- It files a bug report as an “incident” when it should have been a “change” — and the type can’t be corrected afterwards. That’s a new ticket and a merge.
- It tells you it published a knowledge base article. It created an unpublished draft. A month later somebody asks why customers can’t find it.
- It tells you a request was approved. It reads a field that only means “nobody is waiting on a decision” — which is also exactly what rejected looks like.
None of these are the assistant being unintelligent. Every one is a house rule it had no way of knowing.
What it looks like once it’s set up
You describe things in ordinary language. It does the filing.
“The checkout page is dropping card payments when someone retries.”
Files it as an incident, works out the priority from how bad and how widespread it is, puts a short readable title on it, and drops the full detail into the ticket’s history where it belongs.
“What’s on fire right now?”
Pulls a live snapshot of the open queue. Notably, it doesn’t reach for the monthly report — which looks like the right answer and isn’t.
“Close that one out, we fixed it by rolling back the deploy.”
Writes a real explanation into the ticket’s history and into the permanent record on the ticket itself, so somebody opening it in six months sees why it was closed without scrolling through everything.
“Was that change actually approved, and by who?”
Reads the decision itself and tells you who approved it and when — instead of guessing from a field that can’t tell approved from rejected.
What you need
Three things, and two of them you probably already have:
- Claude, with connectors enabled. The skill is instructions — it needs something to give the instructions to.
- A workspace to point it at. New accounts start on a 14-day trial with connector access included, and it doesn’t ask for a card. After that, keeping connector access is $29/month. This skill is written for our platform specifically — if you run a ticketing system somewhere else, it won’t fit it.
- About ten minutes. Most of it is one file you fill in, and your assistant can do a good chunk of that itself.
Setting it up
- Download the file below and unzip it. You’ll get three files: the handbook itself, a setup guide, and a blank profile for your own workspace.
- Install it in Claude the way you install any skill.
- Connect the connector so Claude can actually reach your workspace. The skill is instructions only — without this it does nothing at all.
- Fill in the profile. Open the profile file and ask Claude to “populate the tenant profile from the live connector”. It will look up your teams, people and request types and write most of the file for you.
- Then do the part it can’t. See below — this is the bit that matters.
The ten minutes that actually matter
Your assistant can discover facts about your workspace. It cannot discover your judgement. The profile has a section for exactly that, and it’s the difference between an assistant that files tickets and one that files the right tickets:
- What genuinely counts as urgent here — because “high priority” means something different in every organisation.
- Who signs off on what.
- Which systems nobody touches without asking first.
- Anything that surprised you once. Write it down and it stops costing you time.
Nothing else in the setup repays ten minutes as well as this does.
What it deliberately won’t do
If you’re wary about pointing an AI assistant at real data, this is the important section. These aren’t gaps we haven’t got to. They’re deliberate:
- It won’t approve anything. It brings approvals to you and waits, even when its permissions would technically allow it to decide.
- It won’t publish knowledge base articles. It writes drafts. A person publishes them.
- It won’t text or email people off its own initiative. Messages cost money and reach real humans, so it asks first.
- It can’t delete anything. Every change it makes is reversible.
- It can’t watch your queue. It acts when you ask it to. Nothing gets done while you’re asleep — which is a limitation, and we’d rather you learn it here than discover it during an outage.
Download it
Free, no signup for the file itself, and it contains nothing of ours — no passwords, no server names, no customer data. We check for that before every release.
The individual files are served as plain text, so you can also just
point your assistant at one: “read
https://itsupport.aramagio.com/blog/downloads/52152520e23d2c29/SKILL.md.txt
and install it as a skill.” They download with their proper
.md names.
If you need somewhere to point it, creating a workspace takes about a minute. New accounts get 14 days with connector access included and no card asked for, which is long enough to find out whether this actually fits how you work.
Keeping it current
This version is dated, and it states which parts were re-checked against the live system and which two we could not. We re-issue it whenever the platform moves. If it ever disagrees with what actually happens, that is the document’s fault — tell us and we will fix it.
Practical help desk guides + honest product notes. About one email a week. Unsubscribe anytime.