Set up Skills

What is a Skill?

A Skill is a set of step by step instructions for one specific workflow. It tells Pave Agent what information the analysis needs, what steps to run, and how to lay out the answer.

Skills sit alongside Guidance in Context, and the two do different jobs. Guidance is always-on context about your organization. A Skill applies to one kind of question.

Pave provides a set of built-in Skills that your company loads once, and you can write your own.

How Pave Agent picks a Skill

Pave Agent selects the relevant Skill from your question. You do not need to pick one. Ask your question normally and it works out which Skill applies.

If you already know which Skill you want, type / in the chat box and choose it. Pave Agent then loads that Skill before it starts on your message. Only active Skills appear in the list. See the article on referencing files and Skills in a message.

It reads each Skill's title and description to decide which one fits, so the description is doing real work. Describe the situation the Skill applies to, not just its subject: "identify pay bands that have fallen behind market and quantify the cost to correct them" gives it much more to match on than "band analysis."

Built-in Skills

Pave's built-in Skills cover common compensation workflows, such as benchmarking a role, pricing a new location, evaluating an offer, and investigating a pay complaint. Others help you set up Pave Agent itself, like drafting your Guidance or writing a new Skill.

Pave adds new built-in Skills over time. Once your company has loaded Pave's Skills, new ones are added to your list as they are released. To see the current set, open the Skills tab and look for Pave in the Author column.

Some built-in Skills depend on your own employee data. These are not available to companies that use Pave only for Market Data.

Manage your Skills

1. Open Context from the Pave Agent chat header.
2. Go to the Skills tab.

If the list is empty, select Load Pave Skills to add Pave's built-in Skills. This is a one-time step for your company, and the control disappears once it is done.

From here you can search by title or description, open a Skill to read it, and add a new one. The Author column shows Pave for built-in Skills and your company name for ones you create.

Open a Skill and you can:

• Update its title and description.
• Edit its instructions, using the edit control on the instructions panel. Instructions are edited here rather than in the Update form.
• Turn the Active toggle off to take the Skill out of rotation without deleting it. Turn it back on at any time.
• Delete it permanently. This cannot be undone.

Pave's built-in Skills are yours to change. You can edit one, deactivate it, or delete it outright, the same as a Skill you wrote. If a built-in almost fits your process, edit it rather than starting over. If it does not fit at all, deactivating is the reversible option and worth trying before you delete.

Create your own Skill

The practical way to write one is to ask Pave Agent:

"Help me write a new skill for identifying bands out of sync with market."

It uses the built-in Create a New Skill Skill, which interviews you about the workflow, checks that the analysis you want is something it can actually do with your data, and drafts the Skill. It then offers a dry run as a final check, which it only runs if you confirm.

The draft is copy and paste. Pave Agent cannot save a Skill for you, so copy what it produces and create the Skill yourself on the Skills tab.

To write one from scratch instead, create a Skill and fill in three fields:

• Title: what the workflow is.
• Description: when this Skill should be used. This is what Pave Agent matches your questions against, so be specific about the situation it applies to.
• Instructions: the workflow itself.

What a good Skill contains

The instructions are where most of the value sits. A strong Skill covers:

• What the analysis needs before it can run, and what to ask for if something is missing.
• The steps, broken into phases when the analysis has more than one stage.
• An output template: the exact sections, tables, and columns the answer should come back in, including which caveats to always disclose.
• How to read the result: the common misreadings for this kind of analysis, and how to tell a real problem from an artifact of the data.

The last two are what make a Skill's output consistent enough to forward to someone else without reformatting it first.

Write the instructions in plain compensation language. Skills are read by compensation and HR colleagues, not engineers.

Things to know

Skills are shared across your company. Everyone with Pave Agent access sees the same Skills, and anyone with access can edit or delete them, including the built-in ones.

There is no way to duplicate a Skill. To make a variant, create a new Skill and paste the content in.

A deactivated Skill still exists. It stays in the list and keeps its content, and Pave Agent will not select it until you activate it again.

A Skill changes how Pave Agent approaches a question, not what it can answer from. See the article on what data Pave Agent can see.

You do not choose a Skill

Pave Agent selects the relevant Skill from your question. There is no slash command, no menu, and nothing to pick. Ask your question normally and it works out which Skill applies.

It reads each Skill's title and description to decide which one fits, so the description is doing real work. Describe the situation the Skill applies to, not just its subject: "identify pay bands that have fallen behind market and quantify the cost to correct them" gives it much more to match on than "band analysis."