🐶 Dogear for VSCode

Mark your questions and submit in a batch — inside your editor. Select code, text, or an image file; mix multiline text and reference images in each query; and compose everything for Claude Code or Codex as one prompt. Sibling of the Dogear Chrome extension (in extension/ in the same repo), it shares the same theme, prompt templates, and queue model.

Develop / install

cd vscode
npm install
npm run build     # copies shared files from ../extension and bundles src/

• Run from source: open the vscode/ folder in VSCode and press F5 (Extension Development Host).
• Install for real: npx @vscode/vsce package then code --install-extension dogear-vscode-0.2.1.vsix.

media/theme.js, media/prompts/, and media/fonts/ are build-time copies — the source of truth lives in ../extension/. Edit there, rebuild here.

Usage

1. Choose the context:
   • Select text in any editor → press ⌃⌥Q (Mac) / Ctrl+Alt+Q (Windows/Linux), or right-click → Dogear: Add query for selection.
   • Press the same hotkey without selected text to capture a screen region on local macOS, then ask about the screenshot. On Windows, Linux, and remote extension hosts, Dogear opens the image picker as the fallback.
   • Use Ask tab at the bottom of the sidebar to ask about the active tab without selecting context, or Capture screenshot to start the same screen-region workflow without a hotkey.
   • Right-click a PNG, JPEG, GIF, WebP, BMP, or SVG file in the Explorer or editor title → Dogear: Add query for image.
2. Dogear opens a composer in its sidebar. Enter as many lines as needed, then paste, drop, or choose one or more reference images. The image thumbnails remain inline with the surrounding text and can be reordered. Press Command+Enter / Ctrl+Enter or click Add to queue. Text selections get a highlight; image selections retain a preview in the queued card.
3. Repeat across as many files as you like — the queue is per-workspace and survives restarts.
4. Open the Dogear sidebar (Activity Bar) to edit multiline text and images, reorder inline images, drag cards or file groups by their handles, or delete queries. Click a source header to return to the selected code or image.
5. Deliver the batch:
   • → Codex attaches the queued images through Codex's contributed Add File to Codex Thread command, then pastes the labeled prompt without sending it.
   • → Claude Code pastes the prompt with local image paths because Claude Code currently exposes no file-attachment command to other extensions. Claude can read those files directly; sending remains explicit.
   • As soon as the queue contains an image, a folded Manual handoff tray appears above the delivery buttons and remains available for Copy prompt, → Claude Code, and → Codex. Attach all images to this chat can be used repeatedly, including after switching conversations. When a destination does not expose an attachment command, use the copied local paths because VS Code does not let Dogear inject files into another extension's private webview.

Image bytes live in VS Code's extension storage for the current workspace, not inside the queue JSON or the prompt. Orphaned files are removed when images or questions are deleted. Dogear only gives a destination access when you choose a delivery or copy action.

Markdown previews and AI chat views

• In VS Code's built-in Markdown preview, select rendered text and press ⌃⌥Q / Ctrl+Alt+Q, or right-click → Dogear: Add selected preview text.
• The same right-click capture is available in Codex and Claude Code chats. This webview path briefly uses the clipboard and restores its previous contents immediately.
• VS Code isolates webviews from other extensions, so preview and chat captures are excerpt-only: they can be queued and included in a batch prompt, but Dogear cannot persist a highlight or navigate to the exact rendered passage or chat message.

Highlights re-anchor after edits by exact text + surrounding context, and the line numbers in the composed prompt follow along.

Notes

• The queue lives in the workspace, so questions about repo A never show up while you're in repo B.
• Starting a question from the hotkey always reveals and focuses the Dogear sidebar, even if its webview was already loaded behind another sidebar.
• Neither the Claude Code nor Codex VSCode extensions currently exposes its text composer directly (anthropics/claude-code#27873); Dogear therefore focuses the requested sidebar and uses VS Code's global clipboard command to paste. Sending remains an explicit user action.
• Automated image handoff requires a command contributed by the destination extension. Codex currently provides one; Claude Code does not, so → Claude Code deliberately sends labeled local paths instead of image attachments.
• VS Code's native showInputBox is single-line and has no attachment model, so Dogear's multiline mixed-content input is implemented in its existing sidebar webview.
• VS Code exposes neither screen-region capture nor binary clipboard access to extensions. Dogear therefore uses macOS's interactive screencapture tool locally; other platforms and remote workspaces fall back to choosing an existing screenshot.

Multimodal simulator

After building, run the browser-level sidebar regression harness with:

node vscode/tests/simulator/run.js

See tests/simulator/README.md for its external Playwright dependency. The simulator covers multiline input, three-image attachment, stable queue synchronization, selected-image context, prompt labels, and attach-all manual handoff without writing repository artifacts.