Create AI skills for your team

Not part of the free edition. The AI Skills page under Toolkit is one of the AI administration pages, with AI Usage and AI Activity, and a free organization does not have it. Ask AI itself, and the built-in skills it runs against your business model, are part of the free edition on your own key. See Pricing for what each edition includes.

DeltaVault ships with a catalog of built-in AI skills, ready to run from day one. Your organization is not limited to that catalog. An organization administrator can add a skill of their own to capture something specific to your team, or take a built-in skill and tune it to your own conventions instead of living with the platform default. This guide covers both paths, the dot that marks an org-authored skill everywhere skills appear, and how to publish, retire, or test what you build.

Where org-authored skills live

Org-authored skills live under Toolkit → AI Skills. Open it and you see every skill available to your organization in one rail: the built-in catalog alongside anything your organization has created or customized.

Creating and customizing a skill needs the organization administrator role. Anyone else who opens the page finds it read-only, with a note that an administrator sets skills up. They can still run any published skill from wherever they normally would: the screen they are working on, or Ask AI.

Create a skill for your team

Select New skill and give it two things: a Skill key, the lowercase identifier the rest of the product uses to reference it (letters and digits in segments joined by a period or a hyphen, for example sales.forecast-summary), and a Label, the friendly name your team sees wherever the skill is listed.

A skill key already used by a built-in skill is rejected. That collision is what Customize is for, so DeltaVault points you there instead of letting you shadow a built-in skill by accident. See the next section.

The new skill opens in its editor, on seven tabs:

Identity: the label, description, and icon your team sees.
Where it appears: which surfaces and scopes it runs on, what selection it needs to start, and whether it is offered as a suggested skill (see Control where an AI skill appears).
Prompt: the system prompt and any example exchanges.
Inputs: the form fields a person filling in the skill sees.
Actions: what the skill is allowed to write.
Context: which catalog information and organization conventions it reads.
Settings: its default effort and token budget.

Work through the tabs your skill needs, then select Save in the editor’s toolbar to keep the draft. The editor is a detail panel: your edits wait for Save, Discard changes puts the saved draft back, and picking another skill or tab while the editor holds an unsaved edit asks the Unsaved changes question first. When the draft is ready, publish it from More actions: the command names the version it is about to cut, so it reads Publish v1 for a first release, and while something is missing it stays dimmed and lists what to add (a label, a system prompt, a surface, a scope). Save before you publish: Publish, Test this skill, Repository and Restore a version are dimmed with “Save your changes first.” while the editor holds an unsaved edit. Publishing makes the draft the live version your team runs, and it never takes the skill offline first: the draft stays open for further edits right after.

Customize a built-in skill

Select a built-in skill in the rail and its editor opens read-only, with Customize in More actions beside Test this skill, the only other command a built-in skill offers there. Customize forks the skill’s published content into a new draft that belongs to your organization. From that moment, your organization’s version shadows the built-in one for your team, everywhere that skill appears.

Edit the fork exactly like a skill your organization authored: work through the tabs, select Save, and select Publish from More actions when it is ready. The built-in skill you forked from is untouched. Other organizations keep seeing the original, unless they customize it too.

The org dot

A small blue dot marks every skill your organization added or customized, wherever skills are listed: the rail in Toolkit → AI Skills, and the suggested skills you see in Ask AI. Hover the dot and it reads Added by your organization. A skill with no dot is the platform’s own, unmodified by your organization.

Publish, archive, or remove a customization

Publish, in More actions, makes your current draft the live version your team runs. Publish again whenever you have improved it further.

Retiring a skill depends on how it started:

A skill you created from scratch is retired with Archive, the last command in More actions. It asks “Archive this skill?” first; once you confirm, the skill stops appearing to your team, and its badge then reads archived. There is no unarchive control, so archive only a skill you are sure you are finished with, and create a new one if you need its equivalent later.
A customization of a built-in skill uses Remove customization instead, in the same place. It asks “Remove this customization?” and says that your organization’s version and its published versions are deleted and the platform version takes their place. Once you confirm, your team goes back to seeing the built-in version, exactly as if you had never customized it.

Test before your team sees it

Select Test this skill from More actions and a test panel opens without leaving the page. The command waits while the editor holds an unsaved edit, so save first. Fill in sample inputs and watch the same live progress a real run shows, so you can catch a rough prompt or a missing input before your team runs it for real. A test run is not a free preview: it uses your organization’s AI credits like any other skill run.

AI skills catalog for the full catalog of published skills and what each one produces.
Run an AI skill for how a team member runs a published skill once it is live.