Skip to content

Skills

A skill is an instruction you save once and reuse: how your team writes a weekly summary, the steps for checking a supplier’s portal, the format a report must follow. Instead of pasting the same paragraph into every task, you save it as a skill and add it with one keystroke.

A skill reaches an agent in two ways:

  • You add it to a message. Type / in the composer and pick it. The agent receives your message plus the skill’s text.
  • The agent opens it itself. An agent sees the name and description of the skills it may use, and loads the full text when its instructions or the task call for one. The transcript shows Loading skill … and then Skill … loaded.

Open Skills from the sidebar to see and write them.

In any composer — the New task page, or a task that is already under way — type / to open Agents and skills, and pick one. It sits above the message as a chip until you send; the transcript then records Added the skill … to this task.

A skill can contain blanks: wrap a word in its instructions in braces, such as {customer}, and it becomes a blank. The Skills page lists each skill’s blanks, so you can see what a skill expects before you use it.

Skills are listed in two groups:

GroupWhat is in itWhat you can do
MineThe skills you wroteEdit, share or delete them
Shared by your teamSkills others in the workspace sharedCopy to mine — make your own editable copy

Each row shows the skill’s Name, its Blanks, its Scope, and whether it is Shared or Private. Deleting a skill removes it for everyone; tasks that already used it keep the text they were given.

If you are in more than one workspace, each row also names its Workspace, and a Workspace filter narrows the list to one.

The tools of a browser workflow are skills too, but they belong to their connector: they are not listed here and do not appear in the / menu.

New skill opens a page of its own, with one Save in its header:

SectionWhat it holds
DetailsThe Name (up to 120 characters); the Description (up to 500 — shown beside the name when someone types /, and what an agent reads to decide whether to open it); the Scope, Organization-wide or one named agent; and, only if you are in more than one workspace, the Workspace it belongs to — your main one unless you pick another
InstructionsThe text itself, up to 10,000 characters. {word} becomes a blank
ScriptsCode the steps run on the agent’s computer, the same way every time. Add script gives it a path such as scripts/check.py and its content; scripts sit in folders, and each can be edited, renamed or deleted. When an agent loads the skill, its scripts are placed on the computer. A file under calls/, such as calls/close_ticket.json, is a site call: a request the agent sends from its signed-in browser instead of clicking through the page
HistoryAppears after the first save. Every save is a version: See what changed compares two, and Restore brings an earlier one back as a new version (the name stays). Lock this version keeps one: a locked version shows Locked and cannot be deleted until it is unlocked. Delete removes an unlocked version older than the current one, permanently; you confirm by typing its name, such as Version 3

Someone who can use a skill but not change it sees the page read-only — You can use this skill but not change it. Copy it to make your own. — and gets Copy to mine, which opens a pre-filled new skill; nothing is saved until Save. Leaving a page with unsaved changes asks first.

  • Scoped to an agent, a skill belongs to that agent: it is offered on that agent’s tasks. You can only scope a skill to an agent you can see.
  • Organization-wide, a skill is offered on tasks with no agent, and to any agent it is granted to in the agent builder.

A skill is yours until you share it. Choose Who can use in the skill’s row menu, or the chip in its editor, to open the share box: add people as Can use (they see it and can grant it to an agent they edit) or Can edit, or turn on Everyone in ‹workspace› (an owner or admin). Only the skill’s creator deletes it. See Sharing.

A skill can be granted to an agent in the builder when it is organization-wide and you can use it.

You can also have a skill written for you by teaching the agent the job. In a training session the agent drafts skills from what you tell it, attach and show it — new ones, or improvements to skills it already has, with their scripts and site calls — and lists each one under Learned so far. It can test a skill on its computer before you save it. When you press Save · N, the card shows each skill in full, with a checkbox, and you can edit a skill on the card before you approve it. Approve saves the ticked skills to the agent as its own, together with the other ticked changes, as one new version of the agent.

A skill saved this way belongs to that agent: every task of the agent can load it, those started by an automation included. It stays with that agent until it is shared — see Sharing. The agent builder lists these skills under This agent’s skills. Because it is an ordinary skill, you can read it, edit it and see its versions like any other, with a few differences:

  • It is edited by whoever may edit the agent, not by its author. Anyone else sees it read-only.
  • Its name cannot be changed, and it has no Scope: it belongs to the agent.
  • Saving it writes one new version of the skill and one new version of the agent, so the agent’s Versions record it too.
  • Its History has no Delete. You can see what changed, restore and lock its versions, but not delete one: the agent’s own versions record which version of each skill they used, so none of them is ever removed.
  • Switch off stops every task from loading it until it is switched on again, and Delete removes it from the agent; a task already running keeps what it loaded.

It looks the same wherever you open it: on its page under Skills, or from the agent builder.

In an agent’s builder, Read on a skill — under Skills or This agent’s skills in Configure — opens it in a side panel beside the builder. The panel shows exactly what the skill’s own page shows: Details, Instructions, Scripts (its site calls included) and History, with the page’s actions — Save, Switch off, Share, Delete, whichever you may use — at its foot. On a phone the panel fills the screen and reads like the skill’s page on a phone.

One skill is open at a time. Closing the panel, or opening another skill, with unsaved changes asks first.

What you can change depends on the skill, and the panel follows the same rules as the skill’s page:

SkillWho can edit it thereEveryone else
One of this agent’s own skillsAnyone who may edit the agentSees it read-only
A workspace skill granted to the agentAnyone who may edit that skill: the person who wrote it and people with Can editSees it read-only, with the note A workspace skill given to this agent. It is changed where it is kept, on the Skills page. and Open it on the Skills page

A granted skill from a workspace you are not in is shown read-only, without its History. A read-only panel’s History lists the versions and See what changed, without Restore, Lock or Delete. To work from a skill you cannot change, open it on the Skills page and use Copy to mine.

Some skills ship with AptiveAI Agents and are offered to every agent, beside your workspace’s own. They are not listed on the Skills page, and their text is never shown — the transcript records only that the skill was loaded. Their names all start with aptiveai_, so they do not take a name your own skills might use.

SkillWhat it does
aptiveai_tasksSay where a task stands and give its link, or start a new task — see What an agent can do with your tasks
aptiveai_automationsCreate, change, pause, resume, run or delete an automation, or list them — from a conversation
aptiveai_create_agentTurn the conversation into a new agent of yours, when you ask for one
aptiveai_build_skillSave what was done in the task as a skill, when you ask for one
aptiveai_docsAnswer a how-it-works question about AptiveAI Agents from these docs, with a link to the page it read

A built-in skill can bring tools with it, which a workspace skill cannot: aptiveai_tasks gives the agent tools to look up and start tasks, aptiveai_automations a tool to manage automations, and aptiveai_docs a tool to read this site.

One more built-in skill, aptiveai_site_calls, is used while a site call is being made — in training, Generate with agent and the agent builder — and in a task a person is in, to find out why a site call failed and fix it. It is not offered to a task with no computer, to an automation’s task nobody has written in, or on a workspace that holds regulated data.

ActionWho
Write a skillEveryone in the workspace
Edit a skillThe person who wrote it, and people with Can edit
Delete a skillThe person who wrote it
Share a skill with peopleThe person who wrote it, and people with Can edit
Share it with Everyone in the workspaceOwners and admins
Copy a skill to your ownEveryone who can use it
Edit, switch off or delete an agent’s own skillAnyone who may edit the agent
Share an agent’s own skill with the workspaceAnyone who may edit the agent and is an owner or admin of its workspace
Delete a versionAnyone who may edit the skill — never a version of an agent’s own skill, and never a locked or the current one