Page details
Open a scanned page to curate what the Agent reads, attach Skills to it, and scope where this knowledge applies.
Beta · This feature is currently in beta. Behavior and interface may still evolve.
Click any page card in the Interface list to open its detail panel. Three tabs: Interface, Skills, Details.
Interface tab: what the Agent reads
Trusted Page
The toggle at the top decides the reading mode:
Off: the Agent reads only the elements you allow below. Curated mode.
On: the Agent trusts the whole page and reads it live.
Turning Trusted Page on is equivalent to the Read all mode: anything visible on the page, including user data, becomes readable.
Keep it off for screens that can display sensitive information, and curate below instead.
Elements your agent reads
Elements are grouped by category, each with its count: Navigation, Inputs & Forms, Controls, Data Displays, Contents, Other. Expand a category to see its elements. + Add elements reopens the scan to capture more.
One element in detail
Clicking an element opens its own sheet:
Readable element
The on/off switch. Off excludes the element and any sensitive data it may carry.
About this element
A short description of what the element does, pre-filled by the scan and editable. This is what the Agent relies on to answer.
Selector
The CSS selector locating the element. Selector Strictness is a slider that sets how deep the automatic targeting goes in the CSS nesting: loose matches a broad container, strict pins one exact node. Manual Select re-points the element by hand.
The description is the knowledge. "Navigates to application home/dashboard" tells the Agent exactly what to answer when a user asks what the logo does. Rewrite the pre-filled text wherever your product's vocabulary is more precise.
Skills tab: what the Agent can do here
While the page has no Skills yet, the tab asks What skills do you want to create? and offers the three types, each with a + button.
Once Skills exist, it lists the Enabled Skills attached to this page, each with its type underneath its name (Explain Skill, Guide Skill, Execute Skill). Click one to open its own panel: its usage statistics, and its settings.
You need to be in-app to add skills to an interface.
See Skills and Create a skill.
Two actions sit at the bottom of the panel, whichever tab you are on:
Rescan
Runs the scan again on the page, to pick up elements that changed or appeared since.
Edit in-app
Opens the page in your product with the side panel, to curate elements or add a Skill from the real screen.
Details tab: scope and context
Description for the agent: a summary of what the page shows and what users do on it. The Agent uses it to decide when this page's knowledge is relevant.
URL rules: the environments and URL patterns where this knowledge applies. Add rules manually (Equals, and the other matching operators) or use Generate rules. It relies on the same logic as for targeting experiences via the Where section.
Test agent, at the bottom of the panel, opens a chat scoped to this page's knowledge, the same logic as Test Knowledge.
Related pages
Scan a page: add or re-capture elements
Interface: back to the pages list
Skills: the procedures that run on these elements
Last updated