Turn it on in the dashboard: Assistant › Instructions › Let AI agents use your walkthroughs. It is off by default.
What an agent finds
A public list at https://api.useskate.com/agents/<your site id>.json:
{
"schema": "skate.agents/1",
"site": { "name": "Acme", "url": "https://app.acme.com" },
"about": "Acme uses Skate. These are tasks its team published and verified on the live app. …",
"howTo": ["Open a walkthrough’s link in your browser. Skate opens on the page with a Start card.", "…"],
"walkthroughs": [
{ "id": "wt_0123456789ab", "title": "Invite a teammate", "goal": "Invite a teammate as editor",
"steps": 3, "startsOn": "/members", "mode": "show",
"link": "https://app.acme.com/members#skate=wt_0123456789ab" }
]
}- Only live walkthroughs, at their published version, are listed. Drafts, switched-off ones and ones whose last health check broke are not.
- Nothing about your visitors, usage or account is in it.
- The list answers 404 while the gateway is off or the site is not live.
Your own agents (with an API key) get the same list over MCP: list_walkthroughs and get_walkthrough_link. That works whether or not the public list is on.
An agent already on your page can call Skate.agents(). It returns the same walkthroughs, each with a start() that opens the Start card. It returns [] while the gateway is off.
What it cannot do
The link is the same guide link you paste in a support reply. It opens Skate’s Start card. Nothing else:
- No step runs until someone presses Start or Do it for me.
- Every change waits for Approve, whoever is at the keyboard.
- Skate reports done only when the page shows the result.
- It runs in the browser’s own signed-in session. An agent can only do what that account can do.
- Your “never do” list and show-only mode apply.
Point agents at it
Add a line to your site’s llms.txt or help center:
Tasks you can run on this app with Skate: https://api.useskate.com/agents/<your site id>.json