Week of May 11, 2026

This page describes the interface in May 2026. The current interface uses the comment gutter and keeps tabs in their original order.

New features

PDF and binary file preview. Binary files now open in a built-in viewer instead of trying to render as text. PDFs get a full reader with paging, zoom, and search. See PDFs and other files.
Floating Preview button. When the active tab matches a hook that produces an output file such as a PDF, HTML file, or SVG file, a Preview button appears in the top right of the editor. Click it to open a preview popup that refreshes every time the hook runs. See Live preview.
Find bar. Press Cmd/Ctrl+F to open a find bar for the active tab. Step through matches with Enter / Shift+Enter, and toggle case sensitivity. See Find and appearance.
Image attachments in chat. Drop or paste PNG / JPEG / GIF / WebP images into the chat panel to send them to the agent alongside your message. You can use an attachment to share a screenshot, figure, or reference image. See Send and steer requests.
Drag files in from Finder. Drag files from your filesystem onto the file tree (into any folder) or onto the tab bar to copy them into the workspace and open them. See Files and tabs.
Session viewer. A new viewer lets you replay an agent session from start to finish. It shows tool calls, intermediate messages, and edits, so you can see what the agent did. See Activity and transcript.

Outline pane lists pending edits and comments across all tabs. The outline now surfaces every tab with a pending review round or unseen comment, and clicking an item jumps you straight to that tab and location. This layout was later replaced by the comment gutter and tab badges. See Review edits and Comments and critique.
Pending tabs float next to the active tab. Tabs with pending agent edits are pulled to sit immediately after the active tab so they stay visible without scrolling the tab bar. The current tab bar keeps its original order and shows pending work with badges.
Accept feedback in the editor. Accepting an agent edit now flashes a soft highlight on the accepted text for a moment, so you can identify the text that changed.
Agent plan modal renders markdown. Plans the agent proposes now render with bullets, numbered steps, headings, bold, and inline code instead of as a wall of plaintext. See Plans and long tasks.
Sign in with a Claude Code OAuth token. DocWriter now also accepts CLAUDE_CODE_OAUTH_TOKEN as a credential source, alongside ANTHROPIC_API_KEY, --api-key, and claude login. See Install.

The agent’s Read tool no longer tries to read binary images as text and instead returns a clear error explaining how to attach the image via the chat panel.