Custom Skills

Built-in skills cover general MSP best practices. Custom skills capture your repeatable procedures — the specific multi-step jobs your agents do again and again, like a weekly service-desk report or a monthly license reconciliation. Once one of those jobs works the way you want, you turn it into a skill. From then on, every run follows the exact same steps and produces the same dependable output — instead of the agent figuring the job out from scratch each time.

​

Why custom skills make outputs reliable

When an agent takes on a complex, multi-step task, it normally works out the approach each time it runs — which data to pull, how to crunch it, how to format the result. That flexibility is exactly what you want for one-off problems. But for a job you run every week, you want the opposite: the same proven steps, every time, no surprises. A custom skill gives the agent a fixed recipe to follow instead of reinventing it. The result:

• Consistent — the Monday report looks the same every Monday, in the same shape and format.
• Reliable — the tricky parts (a fiddly query, a specific calculation) were worked out and tested once, then reused. The agent doesn’t re-derive them and occasionally get them wrong.
• Faster — the agent follows the recipe rather than reasoning the whole job out from first principles.

​

What’s inside a skill

A skill is a small bundle of files. There are two kinds: Not every skill has scripts. A skill can be just a SKILL.md — a written procedure the agent reads and follows. Skills that do real data work usually include one or more scripts that the agent runs in its secure code sandbox.

​

Skills that read an outside website

Most skills only talk to the systems Neo already connects to — your PSA, Microsoft 365, your RMM, your documentation. Some need a public web page as well: a vendor status page, a utility’s outage map, a supplier price list. Two separate things control this, and it helps to know which is which:

• The skill records the sites it needs. Neo Support fills this in when the skill is created. It is a statement of what the scripts fetch — nothing more.
• Each agent is allowed to open sites in its own settings. Open the agent, find the Web Fetch tool, and add the site there. This is the permission, and it is per agent, so you stay in control of what each agent can reach.

When an agent loads a skill, Neo compares the two and tells the agent up front if a site the skill needs is not on that agent’s list. You get a clear message naming the site, instead of a failed run that looks like the site is down. So for a skill that reads an outside page:

• Give the address when you ask for the skill, so it is recorded.
• Add that site to the Web Fetch settings of the agent that will run the skill. Until you do, the agent reports the missing site rather than fetching it.
• A saved skill may come back marked “unverified”. Neo Support’s own agent is limited to Neo’s documentation sites and cannot open your vendor’s page, so it saves the skill and tells you which site to allow. That is expected, and the skill is not faulty.

​

How your agents use a skill

• Available everywhere. Every skill you create is automatically available to every agent in your tenant — there’s no per-agent setup or toggle.
• Loaded when relevant. When an agent takes on a task that matches a skill, it loads that skill, reads the procedure, and runs any scripts — instead of working the task out from scratch.
• Guaranteed for a scheduled job. For a scheduled agent whose entire job is one procedure (say, a weekly report), name the skill in the agent’s instructions — for example, “Load and run the weekly-ticket-summary skill” — so it’s used on every run.

​

A typical example

Say every Monday a scheduled agent builds an open-ticket summary for your team — grouped by client and priority, with totals and a short list of the oldest tickets. It works, but you notice small differences week to week: the grouping order shifts, the wording changes, occasionally a count looks off. Turn that job into a skill, and it stops drifting. The procedure and the code are fixed, so the report comes out in the same shape — same sections, same calculations, same format — every single Monday. New technicians reading the report always see the layout they’re used to, and you can trust the numbers because the same tested logic runs each time.

​

Skills vs. instructions & memory

A skill, custom instructions, and memory all shape how an agent behaves, but they do different jobs: The full decision guide — including intents, the fourth piece — is on Teaching Neo.