What a skill is
A skill is a standing instruction your Swantide agent follows whenever the relevant kind of work comes up. A prompt is something you ask once. A skill is how you want the work done every time, without anyone re-explaining it.
Most teams already have these rules. You name custom fields with a department prefix. You always check where a field is used before renaming it. Nobody builds a new report until someone has looked for a duplicate. That knowledge usually sits in one senior admin's head, which means the agent cannot follow it and neither can a new joiner. A skill writes it down once and puts it into the agent.
Where Skills live
In the left navigation, under Knowledge, click Skills.
The page opens on Agent's current skills: a read-only list of everything active in your workspace right now, with a count at the top. Each row shows the skill name, what it does, how often it has been used in the last 30 days, and a badge for where it came from.
That view is read-only on purpose, so anyone can see what the agent is working with. To change anything, click Settings in the top right.
The three places skills come from
Source | Who it applies to | Who can create or change it |
Library | Your whole workspace, once an admin switches one on | Admins can enable, no user can edit as they are written and maintained by Swantide |
Workspace | Your whole workspace | Workspace admins |
Personal | Only you | Anyone |
The Library holds two sets, and a filter at the top of the library view lets you look at each on its own.
Swantide Skill. A small curated set of Salesforce admin practices written by Swantide: check for a duplicate field before creating one, ask which permission sets get access to a new field, run a discovery before building something vague, and similar.
Salesforce Skill. A much larger set that comes from Salesforce's own open-source skills library, published as forcedotcom/sf-skills on GitHub under the Apache 2.0 license. These lean technical and cover the platform broadly: Agentforce, Flow, Apex, LWC, CRM Analytics, Omni-Channel, Data Cloud and more.
You cannot edit a library skill, because you do not own it. If you want one to behave differently, write your own version instead.
Who can do what
Skills settings shows your editing scope in a banner at the top of the page. Check it before you change anything.
Workspace admins switch library skills on and off for the whole team, and create, edit and delete workspace skills. The banner tells you that changes to non-personal skills apply to everyone.
Everyone, admin or not, can create personal skills.
Personal skills belong to you rather than to a workspace, so if you work across several workspaces they follow you into each one.
Turning library skills on and off
Go to Knowledge then Skills, and click Settings.
Click the Library filter, then narrow to Swantide Skill or Salesforce Skill if you want one set at a time.
Find the skill. Use the Search skills box if you know its name, and read the description: it says both when the skill should fire and when it should not.
Flip the On switch. Turn all on and Turn all off act on everything in the current view, which is useful for the Salesforce set if your team does no development work.
Click Done.
Changes take effect on the agent's next turn. You do not need to start a new conversation, though a question already in flight finishes under the previous set.
Creating your own skill
Click + New skill. The first choice in the dialog is who the skill is for:
Just me creates a personal skill. Private to you, and it follows you between workspaces. This is the default.
Everyone in this workspace creates a workspace skill for the whole team. Admins only.
Then fill in two fields: Name, up to 80 characters, and Body, up to 4,000 characters. Click Create, then switch the new skill on.
A list of examples can be found in this article.
A body that works has two parts: when the skill applies, and what to do when it does. Be specific about the trigger, because that first part is what the agent weighs when it decides whether the skill fits the request. Here is the shape:
Apply when creating or renaming a custom field, object, or permission set. Prefix every custom field API name with the owning department code: SLS_ for Sales, SVC_ for Service, FIN_ for Finance. Use PascalCase after the prefix. Labels stay in plain business language with no prefix. If the owning department is not clear from the request, ask before creating anything. |
Keep one rule per skill. A single skill covering naming, testing and documentation is recommended to be broken up into 3 or more skills.
Editing and deleting
Workspace and personal skills carry a pencil and a trash icon on their row. Library skills do not.
Editing a workspace skill changes it for everyone on the next turn, currently there is no version history, so keep a copy of anything substantial before you rewrite it. Deleting is immediate and applies to everyone too.
Calling a skill directly
You do not need to do this for a skill that is switched on. The agent picks those up on its own when they apply. Use this when you want certainty on a particular message.
Type /skillname in the chat box to call directly.
Seeing whether a skill is being used
Every row shows its own usage: how many times it has been used in the last 30 days and when it last fired, or Not used in 30 days.
Use it in both directions. A skill that has never fired is either badly triggered or not needed, and either way it is worth a look. A skill firing far more than you expected is probably matching requests it should not, which is a sign the first line of the body is too broad.
What a skill will not do
Skills shape how the agent works. They never widen what it is allowed to do.
A skill cannot grant access to a feature or a Salesforce permission you do not already have. If the agent cannot take an action for you today, a skill instructing it to will have nothing to act on.
Anything that writes to Salesforce still shows you the plan and waits for your approval. Skills do not change that.
Many skills are read-only by design. They report on the org without touching it.
What On actually means
Turning a skill on makes it available to the agent. It goes into the list the agent can draw on, and the agent decides which entries match the request in front of it. Off means the agent never sees it at all.
So on does not mean the skill shapes every answer, and it does not mean you have to invoke it yourself. An enabled skill gets picked up automatically when the request matches it.
Checking which skills shaped an answer
Because an enabled skill can shape an answer without being asked for, the agent shows which ones it applied in its thinking trace, alongside the rest of its reasoning.
When an answer is not what you expected, look there first. It is usually one of two things: a skill you forgot was on, or one you expected to fire that the agent did not judge relevant. For the second case, call the skill directly with / rather than guessing.
If two enabled skills give conflicting instructions, the agent does not warn you. The trace is how you spot it, and switching one of them off is how you fix it.
