From 0f501d627cb2e83ad4f5ce802acd84a3960da405 Mon Sep 17 00:00:00 2001
From: Phelipe Lorran <112786693+httpsphl@users.noreply.github.com>
Date: Fri, 2 Oct 2026 10:08:04 -0300
Subject: [PATCH 1/2] fix(browser): words built one element per letter keep
 their spaces

The page reader dropped text nodes made of spaces alone. example.com's
script now puts each letter of its English text in its own span, spaces
included, so the bot read "Thisdomainisforuse...". A text node of
spaces is one space now; spaces in a row still collapse.
---
 crates/botloftd/src/browser/reader.js |  5 +++--
 crates/botloftd/tests/browser.rs      | 13 +++++++++++++
 crates/botloftd/tests/common/site.rs  | 11 +++++++++--
 3 files changed, 25 insertions(+), 4 deletions(-)

diff --git a/crates/botloftd/src/browser/reader.js b/crates/botloftd/src/browser/reader.js
index b489f579..46f59c42 100644
--- a/crates/botloftd/src/browser/reader.js
+++ b/crates/botloftd/src/browser/reader.js
@@ -165,8 +165,9 @@
     const walk = (node, parentCursor, depth) => {
       if (++nodes > MAX_NODES || depth > MAX_DEPTH) return;
       if (node.nodeType === Node.TEXT_NODE) {
-        const text = node.nodeValue.replace(/\s+/g, " ");
-        if (text.trim()) out.push(text);
+        // A text node of spaces alone still parts words: pages that put
+        // each letter in its own element keep their spaces that way.
+        out.push(node.nodeValue.replace(/\s+/g, " "));
         return;
       }
       if (node.nodeType !== Node.ELEMENT_NODE) return;
diff --git a/crates/botloftd/tests/browser.rs b/crates/botloftd/tests/browser.rs
index 9b69d4e4..a04195ca 100644
--- a/crates/botloftd/tests/browser.rs
+++ b/crates/botloftd/tests/browser.rs
@@ -89,6 +89,19 @@ async fn a_bot_fills_a_form_after_the_owner_allows_the_site() {
     assert!(closed.expect_err("closed").contains("not open"));
 }
 
+#[tokio::test(flavor = "multi_thread")]
+async fn words_made_of_one_element_per_letter_keep_their_spaces() {
+    let Some(mut b) = setup().await else { return };
+    let opening = call(
+        &b.mcp,
+        "browser_open",
+        json!({ "url": format!("{}/letters", b.site) }),
+    );
+    answer_site(&mut b.app, true, None).await;
+    let page = opening.await.expect("task").expect("page");
+    assert!(page.contains("Two words"), "{page}");
+}
+
 #[tokio::test(flavor = "multi_thread")]
 async fn a_site_the_owner_denies_stays_closed() {
     let Some(mut b) = setup().await else { return };
diff --git a/crates/botloftd/tests/common/site.rs b/crates/botloftd/tests/common/site.rs
index bca71719..d663d1bf 100644
--- a/crates/botloftd/tests/common/site.rs
+++ b/crates/botloftd/tests/common/site.rs
@@ -1,8 +1,9 @@
 //! A small web site on 127.0.0.1 for the browser tests: a form, the page
 //! it leads to, a link that opens a new tab, a button that shows a dialog,
 //! a sign-in page with its fields at fixed points, for the owner's hands,
-//! a page that says how big its window is, one that counts its loads and
-//! one that counts the times it went out of sight.
+//! a page that says how big its window is, one that counts its loads, one
+//! that counts the times it went out of sight and one with each letter in
+//! its own element.
 
 use std::collections::HashMap;
 use std::net::SocketAddr;
@@ -25,6 +26,11 @@ const POPUP: &str = "<!doctype html><title>Links</title>\
     <a href=\"/done?name=Tab&size=Small\" target=\"_blank\">Open in a new tab</a>\
     <button onclick=\"alert('Hi there')\">Warn me</button>";
 
+/// Each letter in a span, spaces too, as example.com's own script does.
+const LETTERS: &str = "<!doctype html><title>Letters</title><p><span>T</span><span>w</span>\
+    <span>o</span><span> </span><span>w</span><span>o</span><span>r</span><span>d</span>\
+    <span>s</span></p>";
+
 /// The fields' middles: user (200, 115), password (200, 215).
 const LOGIN: &str = "<!doctype html><title>Sign in</title><style>body{margin:0}\
     input{position:absolute;left:100px;width:200px;height:30px;margin:0;padding:0;\
@@ -86,6 +92,7 @@ pub async fn serve() -> SocketAddr {
         .route("/login", get(|| async { Html(LOGIN) }))
         .route("/size", get(|| async { Html(SIZE) }))
         .route("/sight", get(|| async { Html(SIGHT) }))
+        .route("/letters", get(|| async { Html(LETTERS) }))
         .route("/account", post(account));
     let listener = TcpListener::bind("127.0.0.1:0").await.expect("bind");
     let addr = listener.local_addr().expect("addr");

From 651abbbd08d05b58382150af8aff36c6649cb8aa Mon Sep 17 00:00:00 2001
From: Phelipe Lorran <112786693+httpsphl@users.noreply.github.com>
Date: Fri, 2 Oct 2026 10:08:20 -0300
Subject: [PATCH 2/2] docs(spec): the page reader keeps a space for a text node
 of spaces

---
 docs/spec.md | 1 +
 1 file changed, 1 insertion(+)

diff --git a/docs/spec.md b/docs/spec.md
index a3388e1b..1cdab4b2 100644
--- a/docs/spec.md
+++ b/docs/spec.md
@@ -1261,6 +1261,7 @@ Esqueceu a senha? [e5 link "Recuperar acesso"]
 ```
 
 - Um script injetado (`Runtime.evaluate`) percorre o documento em ordem de leitura: texto visível, títulos com `#`, itens de lista com `-`, quebras de bloco, e cada controle entre colchetes com uma `ref` (`e<n>`), o papel (`link`, `button`, `textbox`, `checkbox`, `radio`, `select`, `clickable`...), o nome acessível (`aria-label`, `<label>`, texto, `placeholder`, `title`, `alt`) e o estado (valor, marcado, desabilitado). Senhas aparecem só como `(password)`.
+- Um nó de texto só com espaços vira um espaço, e espaços seguidos viram um. Assim uma página que põe cada letra num elemento próprio, espaços incluídos (o script do example.com faz isso com o texto em inglês), não chega ao bot com as palavras grudadas.
 - Entra o que está visível na página inteira, não só na tela: nada com `display: none`, `visibility: hidden`, `aria-hidden` ou tamanho zero. `iframe` do mesmo site e shadow DOM aberto entram; de outro site, só `[frame]`.
 - Elementos clicáveis sem papel (um `div` com `cursor: pointer` que não herda o cursor do pai) viram `clickable`.
 - A mesma ref aponta sempre para o mesmo elemento enquanto ele existir, também entre leituras.