AI skills catalog

An AI skill is a published, ready-to-run capability you launch as an Ask AI tile from the surface you are working on: open the Ask AI panel and the tiles that fit the current screen appear, with its context already loaded. You give a skill a little context, it does focused work against your catalog, and it either returns a proposal you preview and apply, opens a review in the Ask AI pane, or answers in the panel. This page is the catalog of every skill you launch from an Ask AI tile, grouped by the stage of work it serves; each entry says what the skill does, where you launch it, when to reach for it, and what it produces. For the mechanics of running one (watching live progress, reading a failure, scoping which reference documents it reads), see Run an AI skill.

How foundations ground every skill

Before a skill drafts anything, it is grounded in your organization’s foundations. A foundation is shared modeling guidance (house style plus a topic such as Business Modeling, Data Vault Modeling, or Glossary & Naming Standards) that a platform administrator has reviewed and published. The relevant foundations are folded into the skill’s instructions so its proposals speak business language, follow your naming conventions, and respect your chosen modeling method instead of guessing house style each time. Platform administrators create and publish these foundations in the AI Foundations area of the admin settings, where each one pairs your house style with a specific modeling topic.

You do not pick foundations per run; each skill already knows which ones it needs. A model-generation skill draws on the business-modeling and method-specific foundations, an enrichment skill on naming and glossary standards, a quality skill on data-quality-and-profiling guidance. The result is that two people running the same skill on the same catalog get proposals that read consistently, like one team wrote them. For a deeper look at what foundations are and how they reach every run, see the AI Foundations guide.

Catalog a source

Discovery or an import has just landed raw tables, and you need them documented, classified, keyed, and related before anything is modeled on top. These skills work on physical metadata: launch them from the schema explorer, a project or workspace canvas, the catalog Canvas tab, or a connection’s edit page.

Enrich table metadata

Drafts a business-language description for the selected table and its undocumented columns, and proposes a business name plus the matching business entity where one is evident. Launch it from the schema explorer or a project or workspace canvas with a table selected. Run it after discovery lands a raw table, when the metadata is nothing but physical names. It never touches structure: no column, type, or key changes. It produces a proposal you preview and apply; nothing is written until you accept it. See Canvas AI skills for a selected table.

Classify sensitive columns

Judges each column of the selected table for sensitive or regulated data and proposes classifications from your organization’s classification schemes, including any custom schemes you define. Launch it from the schema explorer or a project or workspace canvas with a table selected. Run it as new tables land, so governance coverage starts early instead of trailing the modeling work. It produces one classification proposal per column, which you review and apply. See Canvas AI skills for a selected table.

Analyze schema quality

Reports naming, structure, and data-quality issues across a schema or the tables you select, each finding backed by evidence and rated by severity. Launch it from the schema explorer; no selection is required. Run it before you build models on an unfamiliar schema, or after a discovery run lands new tables, so you know what you are standing on before you build on it. It produces a report and changes nothing.

Analyze impact

Traces which downstream tables, models, and lineage a proposed schema or metadata change would touch, separating certain breakage from possible impact. Launch it from the schema explorer and describe the change you are considering, optionally naming the table it targets. Run it before you rename, retype, or drop anything other work may depend on. It produces a report and changes nothing.

Analyze connection metadata

Reports how well documented a connection’s discovered metadata is: which tables and columns lack descriptions, business names, or classifications, what looks stale against the discovered schema, and what to fix first. Launch it from a connection’s edit page. Run it after a discovery run, so you plan documentation work from evidence instead of guessing where the gaps are. It produces a report and changes nothing.

Discover keys and relationships

Scans the tables on a canvas for primary keys, business keys, and likely physical foreign-key relationships, in any combination of its three passes. Launch it from a project or workspace canvas or the catalog Canvas tab; no selection is needed, the run covers the tables on the canvas, or just the nodes you selected. Run it when discovered or imported tables arrive without keys or joins declared, so both are on record before you design anything that depends on them. A business key lands only on a staging table or a table already in the data vault or data mart layer; scanning a source table redirects its business-key proposal to the staging table it feeds. It opens one review of the ranked proposals, split into a Keys tab and a Relationships tab when you run both passes together, and nothing changes until you accept a proposal; accepting a key replaces the table’s current key. It works at the physical table layer; to propose business relationships or business keys for the entities on a business model, use Discover relationships and keys instead. See Canvas AI skills for tables.

Model the business

With sources documented, you move up a level: domains, models, entities, attributes, and the glossary. These skills launch from the business modeling surface: the models grid or a model’s detail page.

Suggest business domains

Proposes a starter set of top-level business domains with the business models nested under each, grounded in the metadata your organization already tracks. Launch it from the business modeling surface with nothing selected; it works across the whole organization. Run it to scaffold an empty or patchy model catalog rather than naming domains from a blank page. It produces a proposal of domains and models you review; nothing is saved until you accept it.

Suggest business entities

Recommends core business concepts missing from the selected model, each with a category, a one-sentence definition, and a likely business key. Launch it from the business modeling surface with a model selected. Run it after a first modeling pass to close the gaps between what you drew and what the business evidently needs. It produces a proposal of entities you review; nothing is saved until you accept it.

Discover relationships and keys

Two passes you choose with a toggle when you launch it, either alone or together. Suggest relationships proposes business relationships that are missing between the entities on the selected model. Derive business keys looks at every entity on the model that has no designated business key and proposes which of its existing attributes should carry that role, such as an order number or a customer code rather than a surrogate identifier. Launch it from the business modeling surface with a model selected. Reach for it when the entities are on the board but the lines between them, or their business keys, are not. It opens one review of the proposals, relationships and keys grouped separately when you run both passes together, and nothing is saved until you accept them. An entity that already has a designated business key is never changed: the keys pass only fills in what is missing. It works at the business level; to find primary keys, business keys, or foreign-key joins between the physical tables on a canvas, use Discover keys and relationships instead.

Enrich entity metadata

Completes the selected business entity in one pass: a sharper definition, a business key if one is missing, suggested attributes, classifications, and tags. Launch it from the business modeling surface with an entity selected. Reach for it when an entity is little more than a name. It produces a preview of changes that apply only when you accept them.

Enrich attribute descriptions

Drafts a clear business-language description for each attribute on the selected entity that lacks one, leaving good existing descriptions alone. Launch it from the business modeling surface with an entity selected. Run it after generating or importing a model, when attributes carry names but not meaning. It produces a preview of descriptions that apply only when you accept them.

Enrich AI context

Writes or improves the AI context of the selected model, entity, or table and its attributes, so every later AI run starts from better grounding. Launch it from the business modeling surface or the schema explorer with a model, entity, or table selected. Run it to backfill grounding on objects the AI has not yet described; how that context compounds is covered in AI context: self-improving grounding below. Model context is replaced outright; the rest are seeded only where empty. It produces a preview of changes that apply only when you accept them.

Suggest glossary definitions

Drafts a governed glossary definition for each selected entity: one clear business sentence that sets the term apart from its neighbors. Launch it from the business modeling surface with entities selected. Run it to grow the glossary without losing governance. Nothing is saved until you accept, and accepted definitions land as drafts for review.

Review business model

Runs a deep multi-pass review of the selected business model against your chosen requirement documents and the governance baseline. Launch it from the business modeling surface with a model selected. Reach for it before you sign a model off, when you want an independent read instead of one more pass of your own. It is heavyweight: it can take several minutes and it spends real AI credits. It produces a scored report you can reopen later and changes nothing. See Review a business model with AI.

Build the Data Vault

The business model settled, you design the vault: hubs, links, and satellites. All three skills launch from the Data Vault canvas with a vault model open.

Vault canvas assistant

A chat assistant for the Data Vault canvas: ask questions, validate the vault model, and request structural changes in plain language. Launch it from the Data Vault canvas with a vault model open. Reach for it when a change is easier to describe than to draw, or when you want the model explained in Data Vault terms. Every edit it proposes becomes a reviewable action that is applied only after you confirm it.

Validate the Data Vault model

Reviews your Data Vault model for structural faults: hubs without sound business keys, links whose grain mixes units of work, satellites hung off the wrong parent, orphaned constructs, naming drift, and missing source coverage. Launch it from the Data Vault canvas with a vault model open. Run it after accelerating or reshaping a model, while a fault is still cheap to fix on the diagram. It produces a findings report ordered by severity and changes nothing.

Analyze fast-changing attributes

Finds satellite attributes that change far faster than the attributes they are grouped with, using measured profiling: distinct cardinality, row counts, and null ratios. Launch it from the Data Vault canvas with a vault model open. Run it when satellite loads bloat with rows that mostly repeat, the signature of one volatile attribute dragging stable ones along with it. It produces satellite split recommendations grounded in the measurements, and where profiling is missing it says so rather than guessing; it is a report only and changes nothing.

Govern, map, and match

Modeling done, governance carries the work forward: mapping source metadata onto the business model, judging whether sources are fit to load, and keeping your naming and layer conventions in shape.

Check source readiness

Answers one question per source table: is it ready to load? It checks that business keys are declared and mapped to business attributes and that key columns are not nullable, and it reports unmapped column counts and classification coverage as readiness debt. Launch it from the Map & Match workbench and pick the tables to assess. Run it before building loads on a newly landed source, so a missing or nullable key surfaces now rather than in a broken load. It produces a ready or not ready verdict per table, with the evidence behind it; it is a report only and changes nothing.

Profile assistant

Edits your naming, layer, Data Vault, and dimensional profiles from a plain-language request, such as changing a hub prefix or a key-column suffix. Launch it from the profile editors in Settings; no selection is needed. Reach for it when a convention change would otherwise mean hunting through profile fields by hand. Every proposed change is validated and previewed before anything is applied; accepting it edits the profile.

How a skill assembles its prompt

When you run a skill, the model never sees only what you typed. The instructions it receives are assembled from three things, combined for you:

1. Foundations: the published house-style and topical modeling guidance for that skill, so its output follows your organization’s conventions.
2. Context: the relevant slice of your catalog: your organization or project metadata, the table, entity, or model you have selected, and related entities around it. This is gathered automatically from where you launched the skill.
3. Your inputs: the description and pickers you filled in, plus any reference knowledge documents you chose.

Those three layers become the single set of instructions the model works from. It is why a skill’s proposals fit your catalog and your house style without you having to repeat that context every time, and why giving a clear, specific description in the inputs makes the result sharper.

AI context: self-improving grounding

Several skills write more than the object’s name and description: when they generate or enrich a business model, entity, or table, they also author its AI context, a short, business-language brief of the rules, terminology, and relationships a future AI run should know. That context is read back the next time a skill works on the object, so each generation run makes later runs better grounded.

You stay in control of it:

• AI seeds, never overwrites. The AI fills AI context only when it is empty. Once you edit it by hand, the AI leaves your wording alone: your value is authoritative.
• It is read back automatically. When an object is selected or referenced as a picker input, its AI context rides along into the prompt, so you do not re-explain the same rules every run.
• You can backfill it. The Enrich AI context skill authors AI context for the object you have selected: a business model (rewrites the model’s summary), an entity, a table, or an entity’s attributes (seed-only for these). It is safe to re-run; it never overwrites context you have already written.

Keep AI context terse: it is read on every future run that touches the object, so a few high-signal lines beat a paragraph.