Skip to main content
A skill is a named, reusable playbook you hand to the agent: a bundle of know-how plus a hint about which tools to reach for, written for one specific situation. Think “investigate a missing delivery” or “answer a where-is-my-order question.” Each skill captures the steps you’d want followed every time that situation comes up, so the agent doesn’t have to improvise from scratch. Where knowledge is background reference material the agent reads, a skill is a focused procedure the agent pulls up when the moment calls for it. The agent only loads the full procedure when a situation matches — so you can build up a library of detailed playbooks without weighing down every conversation.

​
When to use this

Reach for a skill when you have a recurring situation that should be handled the same considered way each time:
  • A repeatable investigation — the steps your team follows to track down a missing delivery, diagnose a late route, or chase down a status.
  • A standard response pattern — how to answer a customer asking where their order is, or how to walk a return through its stages.
  • A specialized procedure that not every job needs, but that should be exactly right when it does come up.
If you just want to give the agent static context — your SOPs, policies, or definitions — that’s knowledge, not a skill. Knowledge is read as background; a skill is a procedure the agent invokes. Use a skill when there’s a how-to worth pulling up on demand.

​
What’s in a skill

Every skill is described by a few parts. The lightweight parts help the agent (and you) know when to use it; the body is the actual playbook.
PartWhat it is
NameA short, unique handle for the skill (for example, order-status-wismo).
DescriptionA one-line summary of what the skill does.
When to useThe trigger — the situation that should make the agent reach for this skill.
CategoryA grouping label (for example, order tracking or driver operations) used to organize and badge skills.
ContentThe full playbook — the actual step-by-step procedure and know-how the agent follows when it invokes the skill.
Trust tierA label that travels with the skill to signal how cautious its actions are (see the note below).
The trust tier is a label, not a safety control. It surfaces alongside the skill so people can see at a glance how sensitive a procedure is, but it does not by itself force a confirmation or hold an action for approval. The actual guardrails that pause destructive actions for your sign-off live with the action itself — see Custom agents for how high-stakes actions are confirmed. Don’t rely on a skill’s tier as the thing that makes an action safe.

​
How the agent uses a skill

To keep conversations fast, the agent doesn’t carry every skill’s full text around at all times. Instead:
  1. It sees the menu, not the manual. At the start of a conversation, the agent is given the lightweight details of the skills available to it — each skill’s name, description, when-to-use, and category. That’s enough for the agent to recognize when a situation matches a skill.
  2. It pulls the full playbook on demand. When a situation matches, the agent loads that skill’s complete content and follows it. Only the skills it actually needs get pulled in.
  3. You can also invoke a skill directly. If you want to force a specific skill, you can name it with a slash shortcut in chat — for example, typing /order-status-wismo followed by the order reference. This is handy when you know exactly which playbook you want the agent to run.

​
Two kinds of skills

Skills come in two scopes, and the difference is who owns and can edit them.

​
Nash-provided skills

Nash ships a set of ready-made skills that appear for every organization. These cover common operational situations out of the box — things like tracking an order or investigating a delivery problem. They’re maintained by Nash and are read-only to your team: you can see them and your agents can use them, but you can’t edit or delete them.

​
Your organization’s own skills

Your organization can also have its own skills — playbooks authored for the way your team works, available only within your org. These sit alongside the Nash-provided ones, and your agents can use both.
Whether authoring your own org skills is available to you, and exactly who in your org can do it, depends on your account. If you don’t see a way to create or edit skills, that capability may not be turned on for your organization, or it may be reserved for certain roles. Check with whoever administers your Nash account.

​
Handing skills to a custom agent

A custom agent can be told exactly which skills it’s allowed to use. You have three choices:
  • Use the defaults — leave the agent’s skill list unset, and it gets a sensible default set of skills suited to its job.
  • Use none — give the agent an empty skill list to turn skills off for it entirely.
  • Use an explicit list — name the exact skills (by name) the agent may use. The agent then has access to only those, drawn from the Nash-provided skills and your own org skills.
This lets you build a focused agent — for example, a where-is-my-order agent that carries only the order-tracking playbooks it needs, and nothing else.
Pinning an explicit skill list is the reliable way to control what an agent can do. Just like with scope and execution mode, what you set as configuration is enforced on every run, rather than left to the agent’s discretion.

​
FAQ

Knowledge is static reference material the agent reads as background — your SOPs, policies, and links. A skill is a procedure the agent invokes when a situation matches it. Knowledge tells the agent the facts of your operation; a skill tells it the steps to take in a specific scenario. See Knowledge for the reference side.
No. Nash-provided skills are maintained by Nash and are read-only across all organizations. If you want a variant tailored to your team, the path is to create your own org skill (where that’s available to you) rather than editing a Nash one.
At the start of a conversation it’s shown a short menu of available skills — each with a when-to-use trigger. When the situation in front of it matches a trigger, it pulls that skill’s full playbook and follows it. You can also force a specific skill yourself with a slash shortcut in chat.
Not on its own. The trust tier is a label that signals how sensitive a skill is; it doesn’t, by itself, hold an action for your approval. The controls that actually pause destructive actions for confirmation are the per-action guardrails described in Custom agents.
If a custom agent has an explicit skill list, it can use only the skills named there. Leave the list unset to get the default set, or add the skill’s name to the list. Also check that the skill matches the situation — the agent reaches for a skill when the conversation matches its when-to-use trigger.

Custom agents

Hand a curated set of skills to a scoped, reusable agent — and pin exactly what it can do.

Knowledge

Give the agent static reference material — SOPs, policies, and links — as background context.