Using skills: saved expertise your assistant follows

Nobody hands the new analyst a playbook. There isn't one. There is a folder of old decks, two people who know how things are done here, and a set of standards that live in their heads and walk out when they do.

A skill is that playbook, written where your assistant can follow it. Saved expertise: a packaged, repeatable analysis you run by name, with triggers that say when to use it and guardrails that say what it must and must not do.

Quattr ships 39 of them. You keep asking questions in plain language. The skills are why your answers hold their shape from one week to the next, and from one analyst to the next.

A skill has three working parts

Triggers, runs and guardrails.

The guardrails are the part worth studying. The pulse skill cannot do breakdowns or dashboard tours. The significance skills refuse causal language until a test has run, so nothing gets called a cause before a statistical check says the change is bigger than the normal week-to-week wobble.

Refusals live inside the skill rather than in somebody's memory, which means they survive deadline pressure. That is precisely when judgment by feel fails.

None of this asks you to learn a taxonomy. The right skill answers whether or not you know its name. The answer tells you which one ran, so you can inspect it without homework.

The three parts, in the order they fire:

• Triggers: the phrases that summon it. Ask how search is doing and the search-pulse skill picks the question up without you naming it.
• Runs: which pre-built analyses it calls, in what order, and which checks it arms.
• Guardrails: the rules it follows even when a shortcut would be easier.

Why guardrails beat judgment by feel

Because a guardrail does not negotiate, and a person under deadline does.

An assistant without skills answers from general competence, and general competence bends. Ask it leading questions and it follows. Ask a skill a leading question and the guardrail answers instead: the significance check runs first, the scope comes back stated, the metrics that should stay apart stay apart.

That is the difference between an assistant that knows about SEO and one that runs your team's playbook. The first is impressive in a demo. The second is trustworthy on a Tuesday when somebody needs an answer in ten minutes.

The refusals deserve their own appreciation. A skill that cannot sum share percentages, or will not name a cause without a test, does what dashboards never did. It protects you from the answer you asked for.

And it scales with stakes. The higher the meeting an answer is headed for, the more the guardrails matter.

Skills compose into routines

Your daily and weekly checks are skills on a schedule: the same analysis, replayed, so that movement means something.

The workflow card below shows one running as a standing routine rather than a manual chore. It is a worked example of one question answered end to end, with the clicks it replaces.

Composition keeps the assistant's choices sane. Each skill states what it is for and what it refuses, so adjacent skills hand off instead of overlapping. The gallery of all 39 shows each one's triggers, runs and guardrails.

A schedule is also what separates a metric from a screenshot. The same skill run weekly produces comparable answers, and comparable answers produce a trend.

See also: all 39 skills, with their triggers and guardrails →

The workflow that does this: Daily market share vs competitors →

You can read a skill before trusting it

Every skill page shows its job, its trigger phrases, the analyses it runs, and a sample of what it says, including what it declines to say.

That inspectability is the point. Saved expertise you can audit beats a black box that performs confidence at you.

When a skill's answer surprises you, its page is your second stop, after checking where the number came from and what scope it covered. Most surprises resolve into a scope difference or a guardrail doing its job.

The sample outputs are the fastest way in. They show a skill's voice, its verdicts and its refusals, before you have committed anything to a slide.

When no skill fits, that's a finding

A question none of the 39 covers routes to the general analysis layer, which runs the same tools with the same checks armed.

But a question your team keeps asking, with no skill to catch it, is a candidate for a new one. The piece on creating your own Quattr skill covers that path, checks included.

The library is a floor to build on, and what stays non-negotiable sits underneath it. That is the honest answer about life outside the library too. The tools still answer, with the same stated scope and the same refusals, just without a specialist's routine wrapped around them.

See also: writing your own skill, checks included →