Skip WebFetch head metadata text

SqlRush · SqlRush · commit e3b91059d376 · 2026-06-20T20:11:11.000+08:00
diff --git a/docs/cc-100-roadmap.md b/docs/cc-100-roadmap.md
@@ -955,6 +955,8 @@ M7 补充：prompt history `LogEntry` 读取现在接受 `sessionID`/`session`/`
 
 本轮补充：`WebFetch` 文本 body 现在会按 BOM、`Content-Type` charset 或 HTML `<meta charset>`/`http-equiv` charset 解码常见网页编码，包括 UTF-8/UTF-16LE/UTF-16BE、Latin-1 和 Windows-1252，并在 structured content 暴露归一化 `charset`。
 
+本轮补充：`WebFetch` HTML-to-text rendering 现在会先解析 `<base href>` 再跳过 `<head>` 子树，`<title>` 等 head-only metadata 不再污染 rendered body 或 prompt-focused excerpt，同时不破坏相对链接解析。
+
 本轮补充：`WebSearch` JSON parser 现在会递归解包 `web`、`response`、`search`、`hits`、`documents`、`records`、`entries` 等常见搜索后端 wrapper，并继续保留 URL 去重和 allowed/blocked domain filter。
 
 本轮补充：`WebSearch` JSON result parser 现在支持 `pageUrl`/`targetUrl`/`source_url`/`canonicalUrl`/`linkUrl`/`resultUrl`/`destinationUrl`/`clickUrl`/`finalUrl`/`formattedUrl` 等 URL aliases、`htmlTitle`/`htmlSnippet` 等 HTML 标记字段清理、嵌套 URL alias object，以及 `deepLinks`/`siteLinks` 子结果递归解析。
diff --git a/docs/claude-code-go-rewrite-plan.md b/docs/claude-code-go-rewrite-plan.md
@@ -202,6 +202,7 @@ test/parity/                 # golden tests against TS/official behavior
 - 本轮补充：WebFetch HTML-to-text rendering 现在会跳过 `hidden`、`aria-hidden="true"`、CSS `display:none`/`visibility:hidden` 的不可见元素子树，隐藏文本和图片说明不会进入 rendered body 或 prompt-focused excerpt。
 - 本轮补充：WebFetch HTML-to-text rendering 现在会为无可见文本但带 `aria-label`/`title` 的链接保留可访问名称和 resolved href，icon-only 链接可进入 rendered body 与 prompt-focused excerpt。
 - 本轮补充：WebFetch HTML-to-text rendering 现在按浏览器可见性处理 closed `<details>` 和 `<dialog>`：closed details 只渲染第一个 summary 子树，隐藏正文不会进入 excerpt；未带 `open` 的 dialog 会作为不可见子树跳过，open details/dialog 仍正常渲染。
+- 本轮补充：WebFetch HTML-to-text rendering 现在会先解析 `<base href>` 再跳过 `<head>` 子树，`<title>` 等 head-only metadata 不再污染 rendered body 或 prompt-focused excerpt，同时不破坏相对链接解析。
 - 本轮补充：WebSearch HTML 结果解析现在会按搜索页首个有效 `<base href>` 解析相对结果 anchor，覆盖镜像/自定义搜索页中浏览器可见结果 URL 与请求路径不一致的情况。
 - 本轮补充：WebSearch HTML 结果解析现在会读取 `application/ld+json` JSON-LD 结果，递归抽取 `@graph`、`ItemList.itemListElement.item`，支持 JSON-LD `@id` URL alias，并与后续 anchor 结果按 URL 去重。
 - 本轮补充：WebSearch HTML snippet 提取现在识别 Bing 风格 `b_caption`/`b_snippet` 以及常见搜索摘要 class，标题 anchor 后的可见摘要会进入文本输出和 structured result。
diff --git a/internal/tools/web/web_fetch.go b/internal/tools/web/web_fetch.go
@@ -444,6 +444,7 @@ func renderWebFetchBody(contentType string, body string, baseURL string) (string
 	}
 	stripped := removeHTMLWebFetchBlocks(body, "script", "style", "noscript", "template", "svg", "canvas")
 	resolvedBaseURL := webFetchHTMLBaseURL(stripped, baseURL)
+	stripped = removeHTMLWebFetchBlocks(stripped, "head")
 	rendered := stripHTMLWebFetchTags(stripped, resolvedBaseURL)
 	rendered = html.UnescapeString(rendered)
 	return normalizeWebFetchText(rendered), true
diff --git a/internal/tools/web/web_fetch_test.go b/internal/tools/web/web_fetch_test.go
@@ -147,7 +147,7 @@ func TestWebFetchRendersHTMLAndPromptExcerpt(t *testing.T) {
 		_, _ = w.Write([]byte(`<!doctype html>
 <html>
 <head>
-  <title>Plans</title>
+  <title>Invisible beta pricing title leak</title>
   <script>window.secret = "script noise";</script>
   <style>body { color: red; }</style>
 </head>
@@ -185,11 +185,11 @@ func TestWebFetchRendersHTMLAndPromptExcerpt(t *testing.T) {
 	if !ok || !strings.Contains(rendered, "Alpha plan") || !strings.Contains(rendered, "beta pricing plan") {
 		t.Fatalf("rendered body = %#v", result.StructuredContent["rendered_body"])
 	}
-	if strings.Contains(rendered, "script noise") || strings.Contains(rendered, "color: red") {
+	if strings.Contains(rendered, "script noise") || strings.Contains(rendered, "color: red") || strings.Contains(rendered, "Invisible beta pricing title leak") {
 		t.Fatalf("rendered body leaked removed blocks: %#v", rendered)
 	}
 	excerpt, ok := result.StructuredContent["prompt_excerpt"].(string)
-	if !ok || !strings.Contains(excerpt, "beta pricing plan") || strings.Contains(excerpt, "Alpha plan") {
+	if !ok || !strings.Contains(excerpt, "beta pricing plan") || strings.Contains(excerpt, "Alpha plan") || strings.Contains(excerpt, "Invisible beta pricing title leak") {
 		t.Fatalf("prompt excerpt = %#v", result.StructuredContent["prompt_excerpt"])
 	}
 	terms, ok := result.StructuredContent["prompt_terms"].([]string)

Original file line number	Diff line number	Diff line change
`@@ -444,6 +444,7 @@ func renderWebFetchBody(contentType string, body string, baseURL string) (string`
`444`	`444`	`}`
`445`	`445`	`stripped := removeHTMLWebFetchBlocks(body, "script", "style", "noscript", "template", "svg", "canvas")`
`446`	`446`	`resolvedBaseURL := webFetchHTMLBaseURL(stripped, baseURL)`
	`447`	`+ stripped = removeHTMLWebFetchBlocks(stripped, "head")`
`447`	`448`	`rendered := stripHTMLWebFetchTags(stripped, resolvedBaseURL)`
`448`	`449`	`rendered = html.UnescapeString(rendered)`
`449`	`450`	`return normalizeWebFetchText(rendered), true`