v0.1.85

Highlights

• 🤖 Agents can read the browser panel. browser_read gives an agent what you see in the panel: the tab you are looking at (or one whose eye is on) as title, address, visible text, links and, if asked, a screenshot; never the contents of fields, passwords or cookies. Say "read this page" and it works on sites fetch_url cannot reach, like one you are signed in to. What it read counts as outside content, so a sensitive action right after waits for your approval, and the reply lists the page under Sources with a browser mark. browser_open navigates the panel to an address and always asks first, with a card that shows the exact address and whether it goes to the current tab or a new one. Grant both tools to an agent in Agents → tools. Desktop app only.
• 🌐 Browser panel: it never covers the agents rail. With a window too narrow for the width you had chosen, the page view spilled over the agents list beside it. The panel now takes what the chat view can give (at least 360 px, leaving the chat 380 px when there is room) and shrinks and grows with the window.
• 🧩 Skills from GitHub: the button says its state. The row's button reads Update, in green, when the repository folder moved on since you installed the skill, and Up to date, with a check, otherwise; both ask the repository again when pressed, and the answer appears under the row. The manual's Tools chapter explains it.