diff --git a/README.md b/README.md
index db29575..fecfd86 100644
--- a/README.md
+++ b/README.md
@@ -104,6 +104,10 @@ translation, and a machine-readable YAML header — straight into your vault via
 an `obsidian://new` URI. No extra permissions, no server; notes too large for a
 protocol-handler URL fall back to the clipboard.
 
+| One-tap capture, on a real page | The note it writes to your vault |
+| :---: | :---: |
+| ![Capture button in the translate panel on Wikipedia](docs/screenshots/capture-saved.png) | ![The Markdown note OpenRead writes, with its frontmatter handoff contract](docs/screenshots/capture-note.png) |
+
 That header is a deliberate **handoff contract**. Every note is written
 `status: raw`, so a stronger downstream model (a "second brain") can query the
 unprocessed captures, synthesize them, and flip the flag. OpenRead does the
diff --git a/docs/screenshots/capture-note-example.md b/docs/screenshots/capture-note-example.md
new file mode 100644
index 0000000..4573f76
--- /dev/null
+++ b/docs/screenshots/capture-note-example.md
@@ -0,0 +1,22 @@
+---
+title: "Attention Is All You Need"
+source: "Attention Is All You Need — arXiv"
+url: https://arxiv.org/abs/1706.03762
+captured: 2026-07-09T14:05:22.000Z
+from: auto
+to: "Traditional Chinese"
+status: raw
+tags: [openread, transformer, attention, deep-learning]
+---
+
+## Original
+
+> The dominant sequence transduction models are based on complex recurrent or convolutional neural networks that include an encoder and a decoder.
+
+## Translation
+
+主流的序列轉導模型建立在複雜的循環或卷積神經網路之上，這些網路包含一個編碼器與一個解碼器。
+
+## Summary
+
+提出僅以注意力機制構成的 Transformer 架構，完全捨棄循環與卷積。
diff --git a/docs/screenshots/capture-note.png b/docs/screenshots/capture-note.png
new file mode 100644
index 0000000..a74e56d
Binary files /dev/null and b/docs/screenshots/capture-note.png differ
diff --git a/docs/screenshots/capture-saved.png b/docs/screenshots/capture-saved.png
new file mode 100644
index 0000000..b636710
Binary files /dev/null and b/docs/screenshots/capture-saved.png differ