Knowledge lets Chloe answer questions. Skills let her do things: look up an order, file a bug, or check an account. They are what turn an assistant that talks into one that helps.

What a skill is

A job your AI team handles on its own, from start to finish. Each one knows when to start, what to collect, what to do, and when to stop and fetch a person.

Skills are installed rather than built. There is a store of ready-made ones, and you can add them with a click and try them before going live.

Skills, rules and built-in behaviours

Three different things live under Skills, and mixing them up makes the whole area confusing.

Skills: Jobs involving judgement and conversation. Chloe asks questions, checks answers, and decides whether to proceed.
Rules: Instant automations with no conversation and no judgement. Something happens in your inbox and a rule fires: add a label, set priority, route to a team.
Built into Gruvi: Behaviours the platform runs for you before you configure anything, such as response-time targets, asking for a rating, and closing quiet threads.

If the job needs a conversation, it is a skill. If it is "when X happens, always do Y", it is a rule, and you can build one by describing it.

Look through Built into Gruvi early. Several things people assume they must build are already running and only need managing. They include SLA targets, the post-chat survey and auto-close.

The skill store

Skills, then Skill store, is the shelf of ready-made skills. Each carries what it does, whether it takes action or looks things up, and which tool it works with.

Three controls make it navigable: category chips across the top, a search box, and a Works with what you have toggle that hides anything needing a tool you have not connected.

Turn that toggle on first. There is no point reading about a skill that needs a system you do not use.

How to install a skill

1Go to Skills, then Skill store.
2Find one that matches a job your team does over and over.
3Read what it works with, and whether you have that tool connected.
4Click Get.
5It arrives as a draft rather than going live.
6Rehearse it before publishing.

Every skill arrives as a draft. Nothing you install starts acting on customers until you publish it.

What a skill page shows you

An installed skill has its own page with five tabs, and the Overview is written in plain English rather than configuration.

Overview: What this skill does, as a sequence: when it starts, what it collects, what work it does, and how it finishes. Plus what happens if something goes wrong.
Approvals: How much it does before a person steps in.
Rules: Who is allowed to use it, and how much of its reasoning customers may see.
Activity: What it has actually done.
Danger zone: Removing it.

The What if something goes wrong? section is the part worth reading before you publish. It spells out the failure behaviour: what happens if no record is found, if the customer does not confirm, if the action needs sign-off. At the end it says that a person takes over with everything collected, so no customer is left hanging.

Who approves its actions

Under Approvals you choose how much authority the skill has.

Ask for approval: Recommended. The skill prepares each action and asks a person to approve it in the thread. Nothing changes until someone approves.
Approve for me: The skill approves and runs its own actions within limits you set, still asking a person when a limit is hit.

Start on Ask for approval, even for something that looks harmless. Watching a skill prepare twenty actions you then approve teaches you more about whether to trust it than any amount of reading.

Who is allowed to use it

Under Rules, the most consequential setting is whether a customer has to be verified first.

Verified customers only: The customer must have proved who they are: writing from their own email address, being signed in by your app, or entering a code the agent sent them.
Anyone in the chat: Including a visitor who has never signed in and has proved nothing about who they are.

Any skill that reveals account information or changes something should be Verified customers only. "Anyone in the chat" is for genuinely public things, such as looking up public order tracking.

The same tab controls how much of the skill's reasoning a customer may see. It can quote the rules word for word, explain them in its own words, or apply them silently and never mention that rules exist.

Skill suggestions

The Suggestions tab does for skills what gaps do for knowledge: it finds work you should be automating, from what customers actually ask.

Gruvi watches for customers asking for something no skill can finish. Once a day those stalled asks are grouped, and when the same one appears in three or more threads it becomes a suggestion with its fix worked out.

So an empty Suggestions tab is good news, and a busy one is a list of the repetitive work your team is still doing by hand. ---

What skills are and how to install them