Three kinds of skills
- Personal skills — visible only to you. When you create a skill (agent.getswan.com/skills/new) and mark it personal, it lives in your own space (shown in the UI as your personal skills workspace, filterable with
?scope=personal) and can only be used in your own conversations and personal triggers. - Org skills — shared across the whole organization. Any org member can see and use them; they’re the default kind unless you mark a skill personal (filterable with
?scope=org). - System skills — provided by Swan itself, shared across every organization. They’re read-only: you can’t edit or delete them, but Swan can load and follow them. System skills aren’t shown in the Skills page — they work in the background rather than as something you browse. New orgs also ship with a small set of default org skills (for example, a company-qualification skill) that you can see and edit like any other org skill.
Nesting
A skill can have sub-skills nested under it — useful for breaking a large playbook into sections (an overview skill with detailed sub-skills for specific scenarios). Sub-skills aren’t loaded automatically with the parent; Swan fetches them by name or ID when the parent skill’s instructions call for it.Editing and permissions
Editing skills — creating, updating, or deleting any skill (personal or org) — is restricted to Admins by default. Members can view org and personal skills but can’t create or modify them unless an Admin grants that permission (see roles and permissions). Marking a skill “personal” controls who can see and use it, not who can edit it. Every edit to a skill’s instructions is versioned with a change summary, and past versions can be reviewed and restored from the skill’s page in the UI (https://agent.getswan.com/skills/{skillId}).
Skill vs. trigger vs. memory
These three get confused because they all shape agent behavior, but they answer different questions:- A skill is a reusable how-to — instructions for performing a task, invoked by name.
- A trigger is an automation — a source (a schedule, a webhook, a CRM event) plus instructions plus toolkits, running unattended. A trigger’s instructions can reference a skill to reuse existing know-how instead of repeating it.
- Memory is freeform facts, not instructions — notes Swan should remember, not steps it should follow.