Skills
The unit of work LumaVista can be asked to do — invoked from chat, run from a workflow, or applied to a finished report.
A skill is one thing LumaVista knows how to do: extract action items from a transcript, generate a diagram, run a research pass, compute something exactly, send a notification.
Skills matter because they are the same unit everywhere. The thing your workflow calls at 3am is the thing you invoke from chat by name, and the thing offered on a finished report. Not three implementations that drift apart — one, reachable three ways.
Three ways to reach one
From chat. Type / and pick one. It runs against whatever you are looking at.
From a workflow. A workflow step invokes a skill. This is how a scheduled automation does something specific rather than merely prompting a model and hoping.
On a finished report. Some skills apply to work that already exists — generating diagrams for a report is a skill invoked at that moment, with its credit cost shown before it runs.
What ships
| Skill | What it does |
|---|---|
| Extract action items | Pulls commitments out of a transcript and files them. The engine behind Voice to Action. |
| Summarize transcript | Condenses a recording into something readable. |
| Run research | Starts a research pass as a step, rather than as a thing you launch by hand. |
| Research diagram / Diagram enhancer / Research infographic | Turn findings into visuals — see AI-Generated Diagrams. |
| Wolfram Alpha | Computation and reference data, computed rather than recalled. |
| Morning sweep | The scheduled catch-up: what changed, what is waiting. |
| Send notification | Tells you something happened. |
| Workflow designer | Builds workflows conversationally — a skill whose output is automation. |
The catalog is not the interesting part. The extension mechanism is.
Adding your own
Skills are configuration, not code. A skill is a file describing what it does, what it needs, and what it produces. Adding one means adding a file — no deployment, no release, no engineering ticket.
You can upload a skill directly, or describe what you want and have one generated for you, then edit it. Each skill has a detail panel showing what it does and what it is allowed to touch, and its own settings where it takes parameters.
Skill packs bundle related skills so a set arrives together — the note-taker pack installs the capture-and-file chain in one step rather than as four separate decisions. Personas use packs to set up a whole way of working at once; see Personalising LumaVista.
What a skill is allowed to do
A skill is not an escape hatch around the rules the rest of LumaVista follows.
Context is explicit. A skill sees what it has been given, not everything you have. The context toggle on each skill is a boundary, not a hint.
Reaching outside is gated. A skill that contacts an external service is subject to the same send rules as everything else: a person must have seen what it will do. Being invoked from a workflow at 3am does not exempt it — that is precisely the case the rules exist for. See Unattended Actions and Approvals.
Spending is visible. Skills that cost credits show the cost before running, not after.
The point of one unit reachable three ways is that these properties hold in all three. A skill does not become more permissive because a workflow called it instead of you.
Availability. Skills are in beta and off by default. If you have no skills surface and no
/palette in chat, it is not enabled for your account — ask and we will turn it on.
Related
- Building Workflows — where skills run unattended
- Voice to Action — the capture chain skills sit at the end of
- Personalising LumaVista — packs and personas
- Unattended Actions and Approvals — the boundary when a skill reaches outside