send links to the LLM as markdown so drafts keep real hyperlinks

Transcripts flattened links to "text ( url )", so the model copied that shape and drafts came back with the URL in brackets instead of a link. HTML to text for the LLM now writes "[text](url)", which Markdown2HTML turns back into an anchor.
This commit is contained in:
Abhinav Raut
2026-07-31 18:02:22 +05:30
parent da378e7615
commit 2e89586a11
8 changed files with 98 additions and 16 deletions
+1 -1
View File
@@ -661,7 +661,7 @@ func (m *Manager) messageText(msg cmodels.Message) string {
var trimmed string
if msg.ContentType == cmodels.ContentTypeHTML {
// Render full with the same link-keeping options as trimmed, else the prefix diffing below never matches.
if t := stringutil.HTML2TextWithLinks(msg.Content); t != "" {
if t := stringutil.HTML2TextMarkdownLinks(msg.Content); t != "" {
full = t
}
trimmed = stringutil.HTML2TextNoQuotes(msg.Content)
+1 -1
View File
@@ -568,7 +568,7 @@ func Transcript(msgs []Message, max int) string {
}
text := strings.TrimSpace(msg.TextContent)
if msg.ContentType == ContentTypeHTML {
if t := stringutil.HTML2TextWithLinks(msg.Content); t != "" {
if t := stringutil.HTML2TextMarkdownLinks(msg.Content); t != "" {
text = t
}
}
+1 -1
View File
@@ -33,7 +33,7 @@ func TestTranscript(t *testing.T) {
}
got := Transcript(msgs, 50)
want := "Customer: My payment on this page ( https://example.com/pay ) failed.\n" +
want := "Customer: My payment on [this page](https://example.com/pay) failed.\n" +
"Agent: Looking into it.\n" +
"Customer: Any update?\n"
if got != want {
+3 -3
View File
@@ -25,14 +25,14 @@ var (
func HTML2TextNoQuotes(htmlContent string) string {
doc, err := html.Parse(strings.NewReader(htmlContent))
if err != nil {
return TrimPlainTextQuotes(HTML2TextWithLinks(htmlContent))
return TrimPlainTextQuotes(HTML2TextMarkdownLinks(htmlContent))
}
pruneQuotedNodes(doc)
var b strings.Builder
if err := html.Render(&b, doc); err != nil {
return TrimPlainTextQuotes(HTML2TextWithLinks(htmlContent))
return TrimPlainTextQuotes(HTML2TextMarkdownLinks(htmlContent))
}
return TrimPlainTextQuotes(HTML2TextWithLinks(b.String()))
return TrimPlainTextQuotes(HTML2TextMarkdownLinks(b.String()))
}
// TrimPlainTextQuotes strips a trailing quoted-reply block (">" lines, "On ... wrote:" and "Original Message" markers) from plain text.
+1 -1
View File
@@ -41,7 +41,7 @@ func TestHTML2TextNoQuotes(t *testing.T) {
{
name: "protonmail",
html: `<div><div>Hello with quoted.</div><div class="protonmail_signature_block"><div class="protonmail_signature_block-proton">Sent with <a href="https://proton.me/mail/home">Proton Mail</a> secure email.</div></div><div class="protonmail_quote">On Wednesday, July 22nd, 2026 at 9:06 PM, Abhinav &lt;user@example.com&gt; wrote:<br><blockquote class="protonmail_quote" type="cite"><p>Hello!</p></blockquote><br></div></div>`,
want: "Hello with quoted.\nSent with Proton Mail ( https://proton.me/mail/home ) secure email.",
want: "Hello with quoted.\nSent with [Proton Mail](https://proton.me/mail/home) secure email.",
},
{
name: "quote only returns empty",
+67
View File
@@ -0,0 +1,67 @@
package stringutil
import (
"strings"
"github.com/jaytaylor/html2text"
"golang.org/x/net/html"
)
// HTML2TextMarkdownLinks converts HTML to text keeping links as markdown "[text](url)", the form a
// model copies back into its reply and that Markdown2HTML turns into an anchor again.
func HTML2TextMarkdownLinks(htmlContent string) string {
if doc, err := html.Parse(strings.NewReader(htmlContent)); err == nil {
inlineAnchorMarkdown(doc)
var b strings.Builder
if err := html.Render(&b, doc); err == nil {
htmlContent = b.String()
}
}
return htmlToText(htmlContent, html2text.Options{})
}
func inlineAnchorMarkdown(n *html.Node) {
var next *html.Node
for child := n.FirstChild; child != nil; child = next {
next = child.NextSibling
if child.Type == html.ElementNode && child.Data == "a" {
link := markdownLink(nodeText(child), strings.TrimSpace(attrValue(child, "href")))
if link != "" {
n.InsertBefore(&html.Node{Type: html.TextNode, Data: link}, child)
n.RemoveChild(child)
continue
}
}
inlineAnchorMarkdown(child)
}
}
// markdownLink returns "" when the anchor is not worth rewriting, leaving it to the text conversion.
func markdownLink(text, href string) string {
if href == "" || strings.HasPrefix(href, "#") || strings.HasPrefix(href, "cid:") {
return ""
}
text = strings.Join(strings.Fields(text), " ")
if text == "" || text == href {
return href
}
if strings.ContainsAny(href, " ()") {
href = "<" + href + ">"
}
return "[" + strings.NewReplacer("[", `\[`, "]", `\]`).Replace(text) + "](" + href + ")"
}
func nodeText(n *html.Node) string {
var b strings.Builder
var walk func(*html.Node)
walk = func(node *html.Node) {
if node.Type == html.TextNode {
b.WriteString(node.Data)
}
for c := node.FirstChild; c != nil; c = c.NextSibling {
walk(c)
}
}
walk(n)
return b.String()
}
-5
View File
@@ -48,11 +48,6 @@ func HTML2Text(html string) string {
return htmlToText(html, html2text.Options{TextOnly: true})
}
// HTML2TextWithLinks converts HTML to text keeping link URLs as "text ( url )".
func HTML2TextWithLinks(html string) string {
return htmlToText(html, html2text.Options{})
}
// Markdown2HTML converts markdown to HTML, falling back to the input on error.
func Markdown2HTML(md string) string {
var b strings.Builder
+24 -4
View File
@@ -297,16 +297,16 @@ func TestSplitName(t *testing.T) {
}
}
func TestHTML2TextWithLinks(t *testing.T) {
func TestHTML2TextMarkdownLinks(t *testing.T) {
tests := []struct {
name string
html string
want string
}{
{
name: "link with distinct text keeps url",
name: "link with distinct text becomes a markdown link",
html: `<p>See <a href="https://example.com/guide">the guide</a> for steps.</p>`,
want: "See the guide ( https://example.com/guide ) for steps.",
want: "See [the guide](https://example.com/guide) for steps.",
},
{
name: "link text equal to url not duplicated",
@@ -318,10 +318,30 @@ func TestHTML2TextWithLinks(t *testing.T) {
html: `<p>No links here.</p>`,
want: "No links here.",
},
{
name: "nested markup inside the anchor flattens to link text",
html: `<p><a href="https://example.com/x"><strong>Pay</strong> now</a></p>`,
want: "[Pay now](https://example.com/x)",
},
{
name: "brackets in link text are escaped",
html: `<p><a href="https://example.com">Docs [beta]</a></p>`,
want: `[Docs \[beta\]](https://example.com)`,
},
{
name: "url with parentheses is wrapped in angle brackets",
html: `<p><a href="https://example.com/a(b)">See</a></p>`,
want: "[See](<https://example.com/a(b)>)",
},
{
name: "anchor without href keeps its text",
html: `<p><a name="top">Top</a> of page.</p>`,
want: "Top of page.",
},
}
for _, tt := range tests {
t.Run(tt.name, func(t *testing.T) {
if got := HTML2TextWithLinks(tt.html); got != tt.want {
if got := HTML2TextMarkdownLinks(tt.html); got != tt.want {
t.Errorf("got %q, want %q", got, tt.want)
}
})