Picking the Right Job for a Custom Skill

You have installed a skill, run its setup interview, and bent it to your voice. Time to build one from scratch. Before you write a single line, though, you have one decision that matters more than anything you type later: which job gets the skill. Most homemade skills that end up abandoned were doomed at this step, not at the writing step.

The ten-minute task audit

Grab a piece of paper or open a note. List every task you personally did three or more times last month. Not the big projects, the recurring stuff: quotes you priced, updates you emailed, reviews you answered, invoices you chased. For a typical owner-operator that list runs eight to fifteen items.

Now score each one from 1 to 5 on three dimensions:

Repetition earns back your build time. Stable context is what a skill file stores, so a task that changes its rules weekly scores low. Pain is your motivation to keep the skill maintained after week two. Add up the three numbers. Your highest total is your candidate.

Then apply the sizing test: pick a job, not a department. "Write my quotes from site notes" is a job with clear inputs and one output. "Run my marketing" is a department, and a skill asked to do a department's work produces vague output on every task inside it. If your winner sounds like something you would put in a job posting title, shrink it until it sounds like a bullet point inside one.

Good first skills, and bad ones

Three first skills we see work over and over in client businesses:

• Quoting. Same services, same rate card, same structure, new customer every time. Perfect stability.
• Weekly client updates. Say you run a two-person renovation outfit with four active jobs. Every Friday you owe four updates in the same format with different progress notes. That is repetition and stable context in one package.
• Social replies in your voice. Comments and reviews arrive daily, your tone rules stay fixed, and the writing tax is constant. A skill loaded with your voice and a few real examples handles the drafting while you keep the final say.

Two categories to avoid for a first build:

• Anything too broad. "Handle my customers," "do my admin," "run my marketing." Broad skills have no stable context to store, so they collapse back into generic AI output, the thing you took this course to escape.
• Anything where being wrong is expensive and nobody reviews the output. Contracts, tax filings, anything regulatory. An assistant can still state wrong things with full confidence, and a skill does not change that. If a mistake in the output could cost you a client, a fine, or a lawsuit, that task either stays human or gets a mandatory review step built into your process. Finance and legal drafts always go past a professional before they bind you to anything.

That last point is the honest limit of this whole chapter: a skill drafts, you decide. Even the best custom skill earns a read-through before its output reaches a customer.

Check the toolbox before you build

One more gate before you start writing. Search the Vexlo toolbox for your task first. There are 20 free skills covering the most common small-business jobs, and customizing an existing one takes a fraction of the effort of building from a blank file. If your winner is quoting or proposals, Proposal Studio already exists and Chapter 2 taught you how to adapt it. Support replies are covered by the customer support autopilot, and review responses by the review and reputation manager.

Build from scratch when your task is specific to how your business runs: your quoting logic, your update format, your voice. That specificity is exactly what makes a custom skill worth owning, and it is what the next lesson builds, section by section.

Try it now

Run the audit tonight. List last month's repeated tasks, score each on repetition, stable context, and pain, and circle the single winner. Then spend five minutes in the toolbox confirming nobody has built it for you already. Bring the winner to the next lesson; you will write its skill file there.