gmail: strip Outlook reply headers without inline <b>/<strong> wrappers

connectors/gmail/src/gmail-api.ts

@@ -438,12 +438,37 @@ function extractBody(part: GmailMessagePart): { content: string; contentType: "t
  return { content: "", contentType: "text" };
 }
 
+/**
+ * Locates the start of an Outlook-style "From: / Sent: / To: / Subject:"
+ * reply header even when the field labels are not wrapped in `<b>` or
+ * `<strong>` — e.g. corporate Exchange / Outlook variants that put the
+ * label in a `<span style="font-weight:bold">`, a `<font>` tag, or a
+ * plain `MsoNormal` paragraph with no inline bold at all.
+ *
+ * Strategy: replace every HTML tag with a same-length run of spaces so
+ * character offsets in the stripped text still map 1:1 back to the
+ * original. Then require each label to start at a structural boundary
+ * (start of string, a real newline, or 3+ whitespace chars — the smallest
+ * gap any HTML block tag produces when replaced). That anchor is what
+ * keeps user-written prose from false-matching.
+ *
+ * Returns the index of "From:" in the original content, or -1 if no
+ * Outlook reply header is found.
+ */
+export function findOutlookHeaderTagAgnostic(content: string): number {
+ const flat = content.replace(/<[^>]*>/g, (m) => " ".repeat(m.length));
+ const re =
+ /(?<=^|\n|[ \t]{3,})From:[\s\S]{0,1500}?(?<=\n|[ \t]{3,})Sent:[\s\S]{0,800}?(?<=\n|[ \t]{3,})To:[\s\S]{0,1500}?(?<=\n|[ \t]{3,})Subject:/i;
+ const m = flat.match(re);
+ return m?.index ?? -1;
+}
+
 /**
  * Strips quoted reply content from an email body.
  * Since Plot shows each message as a separate note in a thread,
  * the quoted previous messages are redundant noise.
  */
-function stripQuotedReply(
+export function stripQuotedReply(
  content: string,
  contentType: "text" | "html"
 ): string {
@@ -477,16 +502,18 @@ function stripQuotedReply(
  }
 
  // Outlook (desktop, OWA, and corporate Exchange clients) wraps replies
- // with a bolded "From: / Sent: / To: / Subject:" header block. The
- // markup around it varies — sometimes a <div style="border-top:..."
- // divider, sometimes <hr>, sometimes neither — so detect the stable
- // header sequence first, then walk back to the nearest structural
- // boundary.
+ // with a "From: / Sent: / To: / Subject:" header block. The markup
+ // varies — sometimes `<b>` or `<strong>`, sometimes `<span
+ // style="font-weight:bold">`, sometimes a `MsoNormal` paragraph with
+ // no inline bold at all (Gowling-style corporate Exchange). Try the
+ // tight bold-wrapped pattern first, then fall back to a tag-agnostic
+ // boundary match.
  const outlookHeaderRe =
  /<(b|strong)[^>]*>\s*From:?\s*<\/\1>[\s\S]{0,1000}<(b|strong)[^>]*>\s*Sent:?\s*<\/\2>[\s\S]{0,1000}<(b|strong)[^>]*>\s*To:?\s*<\/\3>[\s\S]{0,1000}<(b|strong)[^>]*>\s*Subject:?\s*<\/\4>/i;
  const outlookHeaderMatch = content.match(outlookHeaderRe);
- if (outlookHeaderMatch?.index !== undefined) {
- const fromIdx = outlookHeaderMatch.index;
+ const fromIdx =
+ outlookHeaderMatch?.index ?? findOutlookHeaderTagAgnostic(content);
+ if (fromIdx !== -1) {
  const lookbackStart = Math.max(0, fromIdx - 1000);
  const lookback = content.substring(lookbackStart, fromIdx);
  // Prefer the latest structural divider (border-top div or <hr>)

connectors/gmail/src/gmail-api.ts

@@ -438,12 +438,37 @@ function extractBody(part: GmailMessagePart): { content: string; contentType: "t
  return { content: "", contentType: "text" };
 }
 
+/**
+ * Locates the start of an Outlook-style "From: / Sent: / To: / Subject:"
+ * reply header even when the field labels are not wrapped in `<b>` or
+ * `<strong>` — e.g. corporate Exchange / Outlook variants that put the
+ * label in a `<span style="font-weight:bold">`, a `<font>` tag, or a
+ * plain `MsoNormal` paragraph with no inline bold at all.
+ *
+ * Strategy: replace every HTML tag with a same-length run of spaces so
+ * character offsets in the stripped text still map 1:1 back to the
+ * original. Then require each label to start at a structural boundary
+ * (start of string, a real newline, or 3+ whitespace chars — the smallest
+ * gap any HTML block tag produces when replaced). That anchor is what
+ * keeps user-written prose from false-matching.
+ *
+ * Returns the index of "From:" in the original content, or -1 if no
+ * Outlook reply header is found.
+ */
+export function findOutlookHeaderTagAgnostic(content: string): number {
+ const flat = content.replace(/<[^>]*>/g, (m) => " ".repeat(m.length));
+ const re =
+ /(?<=^|\n|[ \t]{3,})From:[\s\S]{0,1500}?(?<=\n|[ \t]{3,})Sent:[\s\S]{0,800}?(?<=\n|[ \t]{3,})To:[\s\S]{0,1500}?(?<=\n|[ \t]{3,})Subject:/i;
+ const m = flat.match(re);
+ return m?.index ?? -1;
+}
+
 /**
  * Strips quoted reply content from an email body.
  * Since Plot shows each message as a separate note in a thread,
  * the quoted previous messages are redundant noise.
  */
-function stripQuotedReply(
+export function stripQuotedReply(
  content: string,
  contentType: "text" | "html"
 ): string {
@@ -477,16 +502,18 @@ function stripQuotedReply(
  }
 
  // Outlook (desktop, OWA, and corporate Exchange clients) wraps replies
- // with a bolded "From: / Sent: / To: / Subject:" header block. The
- // markup around it varies — sometimes a <div style="border-top:..."
- // divider, sometimes <hr>, sometimes neither — so detect the stable
- // header sequence first, then walk back to the nearest structural
- // boundary.
+ // with a "From: / Sent: / To: / Subject:" header block. The markup
+ // varies — sometimes `<b>` or `<strong>`, sometimes `<span
+ // style="font-weight:bold">`, sometimes a `MsoNormal` paragraph with
+ // no inline bold at all (Gowling-style corporate Exchange). Try the
+ // tight bold-wrapped pattern first, then fall back to a tag-agnostic
+ // boundary match.
  const outlookHeaderRe =
  /<(b|strong)[^>]*>\s*From:?\s*<\/\1>[\s\S]{0,1000}<(b|strong)[^>]*>\s*Sent:?\s*<\/\2>[\s\S]{0,1000}<(b|strong)[^>]*>\s*To:?\s*<\/\3>[\s\S]{0,1000}<(b|strong)[^>]*>\s*Subject:?\s*<\/\4>/i;
  const outlookHeaderMatch = content.match(outlookHeaderRe);
- if (outlookHeaderMatch?.index !== undefined) {
- const fromIdx = outlookHeaderMatch.index;
+ const fromIdx =
+ outlookHeaderMatch?.index ?? findOutlookHeaderTagAgnostic(content);
+ if (fromIdx !== -1) {
  const lookbackStart = Math.max(0, fromIdx - 1000);
  const lookback = content.substring(lookbackStart, fromIdx);
  // Prefer the latest structural divider (border-top div or <hr>)

connectors/gmail/src/gmail-api.ts

@@ -438,12 +438,37 @@ function extractBody(part: GmailMessagePart): { content: string; contentType: "t
  return { content: "", contentType: "text" };
 }
 
+/**
+ * Locates the start of an Outlook-style "From: / Sent: / To: / Subject:"
+ * reply header even when the field labels are not wrapped in `<b>` or
+ * `<strong>` — e.g. corporate Exchange / Outlook variants that put the
+ * label in a `<span style="font-weight:bold">`, a `<font>` tag, or a
+ * plain `MsoNormal` paragraph with no inline bold at all.
+ *
+ * Strategy: replace every HTML tag with a same-length run of spaces so
+ * character offsets in the stripped text still map 1:1 back to the
+ * original. Then require each label to start at a structural boundary
+ * (start of string, a real newline, or 3+ whitespace chars — the smallest
+ * gap any HTML block tag produces when replaced). That anchor is what
+ * keeps user-written prose from false-matching.
+ *
+ * Returns the index of "From:" in the original content, or -1 if no
+ * Outlook reply header is found.
+ */
+export function findOutlookHeaderTagAgnostic(content: string): number {
+ const flat = content.replace(/<[^>]*>/g, (m) => " ".repeat(m.length));
+ const re =
+ /(?<=^|\n|[ \t]{3,})From:[\s\S]{0,1500}?(?<=\n|[ \t]{3,})Sent:[\s\S]{0,800}?(?<=\n|[ \t]{3,})To:[\s\S]{0,1500}?(?<=\n|[ \t]{3,})Subject:/i;
+ const m = flat.match(re);
+ return m?.index ?? -1;
+}
+
 /**
  * Strips quoted reply content from an email body.
  * Since Plot shows each message as a separate note in a thread,
  * the quoted previous messages are redundant noise.
  */
-function stripQuotedReply(
+export function stripQuotedReply(
  content: string,
  contentType: "text" | "html"
 ): string {
@@ -477,16 +502,18 @@ function stripQuotedReply(
  }
 
  // Outlook (desktop, OWA, and corporate Exchange clients) wraps replies
- // with a bolded "From: / Sent: / To: / Subject:" header block. The
- // markup around it varies — sometimes a <div style="border-top:..."
- // divider, sometimes <hr>, sometimes neither — so detect the stable
- // header sequence first, then walk back to the nearest structural
- // boundary.
+ // with a "From: / Sent: / To: / Subject:" header block. The markup
+ // varies — sometimes `<b>` or `<strong>`, sometimes `<span
+ // style="font-weight:bold">`, sometimes a `MsoNormal` paragraph with
+ // no inline bold at all (Gowling-style corporate Exchange). Try the
+ // tight bold-wrapped pattern first, then fall back to a tag-agnostic
+ // boundary match.
  const outlookHeaderRe =
  /<(b|strong)[^>]*>\s*From:?\s*<\/\1>[\s\S]{0,1000}<(b|strong)[^>]*>\s*Sent:?\s*<\/\2>[\s\S]{0,1000}<(b|strong)[^>]*>\s*To:?\s*<\/\3>[\s\S]{0,1000}<(b|strong)[^>]*>\s*Subject:?\s*<\/\4>/i;
  const outlookHeaderMatch = content.match(outlookHeaderRe);
- if (outlookHeaderMatch?.index !== undefined) {
- const fromIdx = outlookHeaderMatch.index;
+ const fromIdx =
+ outlookHeaderMatch?.index ?? findOutlookHeaderTagAgnostic(content);
+ if (fromIdx !== -1) {
  const lookbackStart = Math.max(0, fromIdx - 1000);
  const lookback = content.substring(lookbackStart, fromIdx);
  // Prefer the latest structural divider (border-top div or <hr>)

connectors/gmail/src/gmail-api.ts

@@ -438,12 +438,37 @@ function extractBody(part: GmailMessagePart): { content: string; contentType: "t
  return { content: "", contentType: "text" };
 }
 
+/**
+ * Locates the start of an Outlook-style "From: / Sent: / To: / Subject:"
+ * reply header even when the field labels are not wrapped in `<b>` or
+ * `<strong>` — e.g. corporate Exchange / Outlook variants that put the
+ * label in a `<span style="font-weight:bold">`, a `<font>` tag, or a
+ * plain `MsoNormal` paragraph with no inline bold at all.
+ *
+ * Strategy: replace every HTML tag with a same-length run of spaces so
+ * character offsets in the stripped text still map 1:1 back to the
+ * original. Then require each label to start at a structural boundary
+ * (start of string, a real newline, or 3+ whitespace chars — the smallest
+ * gap any HTML block tag produces when replaced). That anchor is what
+ * keeps user-written prose from false-matching.
+ *
+ * Returns the index of "From:" in the original content, or -1 if no
+ * Outlook reply header is found.
+ */
+export function findOutlookHeaderTagAgnostic(content: string): number {
+ const flat = content.replace(/<[^>]*>/g, (m) => " ".repeat(m.length));
+ const re =
+ /(?<=^|\n|[ \t]{3,})From:[\s\S]{0,1500}?(?<=\n|[ \t]{3,})Sent:[\s\S]{0,800}?(?<=\n|[ \t]{3,})To:[\s\S]{0,1500}?(?<=\n|[ \t]{3,})Subject:/i;
+ const m = flat.match(re);
+ return m?.index ?? -1;
+}
+
 /**
  * Strips quoted reply content from an email body.
  * Since Plot shows each message as a separate note in a thread,
  * the quoted previous messages are redundant noise.
  */
-function stripQuotedReply(
+export function stripQuotedReply(
  content: string,
  contentType: "text" | "html"
 ): string {
@@ -477,16 +502,18 @@ function stripQuotedReply(
  }
 
  // Outlook (desktop, OWA, and corporate Exchange clients) wraps replies
- // with a bolded "From: / Sent: / To: / Subject:" header block. The
- // markup around it varies — sometimes a <div style="border-top:..."
- // divider, sometimes <hr>, sometimes neither — so detect the stable
- // header sequence first, then walk back to the nearest structural
- // boundary.
+ // with a "From: / Sent: / To: / Subject:" header block. The markup
+ // varies — sometimes `<b>` or `<strong>`, sometimes `<span
+ // style="font-weight:bold">`, sometimes a `MsoNormal` paragraph with
+ // no inline bold at all (Gowling-style corporate Exchange). Try the
+ // tight bold-wrapped pattern first, then fall back to a tag-agnostic
+ // boundary match.
  const outlookHeaderRe =
  /<(b|strong)[^>]*>\s*From:?\s*<\/\1>[\s\S]{0,1000}<(b|strong)[^>]*>\s*Sent:?\s*<\/\2>[\s\S]{0,1000}<(b|strong)[^>]*>\s*To:?\s*<\/\3>[\s\S]{0,1000}<(b|strong)[^>]*>\s*Subject:?\s*<\/\4>/i;
  const outlookHeaderMatch = content.match(outlookHeaderRe);
- if (outlookHeaderMatch?.index !== undefined) {
- const fromIdx = outlookHeaderMatch.index;
+ const fromIdx =
+ outlookHeaderMatch?.index ?? findOutlookHeaderTagAgnostic(content);
+ if (fromIdx !== -1) {
  const lookbackStart = Math.max(0, fromIdx - 1000);
  const lookback = content.substring(lookbackStart, fromIdx);
  // Prefer the latest structural divider (border-top div or <hr>)

connectors/gmail/src/gmail-api.ts

@@ -438,12 +438,37 @@ function extractBody(part: GmailMessagePart): { content: string; contentType: "t
  return { content: "", contentType: "text" };
 }
 
+/**
+ * Locates the start of an Outlook-style "From: / Sent: / To: / Subject:"
+ * reply header even when the field labels are not wrapped in `<b>` or
+ * `<strong>` — e.g. corporate Exchange / Outlook variants that put the
+ * label in a `<span style="font-weight:bold">`, a `<font>` tag, or a
+ * plain `MsoNormal` paragraph with no inline bold at all.
+ *
+ * Strategy: replace every HTML tag with a same-length run of spaces so
+ * character offsets in the stripped text still map 1:1 back to the
+ * original. Then require each label to start at a structural boundary
+ * (start of string, a real newline, or 3+ whitespace chars — the smallest
+ * gap any HTML block tag produces when replaced). That anchor is what
+ * keeps user-written prose from false-matching.
+ *
+ * Returns the index of "From:" in the original content, or -1 if no
+ * Outlook reply header is found.
+ */
+export function findOutlookHeaderTagAgnostic(content: string): number {
+ const flat = content.replace(/<[^>]*>/g, (m) => " ".repeat(m.length));
+ const re =
+ /(?<=^|\n|[ \t]{3,})From:[\s\S]{0,1500}?(?<=\n|[ \t]{3,})Sent:[\s\S]{0,800}?(?<=\n|[ \t]{3,})To:[\s\S]{0,1500}?(?<=\n|[ \t]{3,})Subject:/i;
+ const m = flat.match(re);
+ return m?.index ?? -1;
+}
+
 /**
  * Strips quoted reply content from an email body.
  * Since Plot shows each message as a separate note in a thread,
  * the quoted previous messages are redundant noise.
  */
-function stripQuotedReply(
+export function stripQuotedReply(
  content: string,
  contentType: "text" | "html"
 ): string {
@@ -477,16 +502,18 @@ function stripQuotedReply(
  }
 
  // Outlook (desktop, OWA, and corporate Exchange clients) wraps replies
- // with a bolded "From: / Sent: / To: / Subject:" header block. The
- // markup around it varies — sometimes a <div style="border-top:..."
- // divider, sometimes <hr>, sometimes neither — so detect the stable
- // header sequence first, then walk back to the nearest structural
- // boundary.
+ // with a "From: / Sent: / To: / Subject:" header block. The markup
+ // varies — sometimes `<b>` or `<strong>`, sometimes `<span
+ // style="font-weight:bold">`, sometimes a `MsoNormal` paragraph with
+ // no inline bold at all (Gowling-style corporate Exchange). Try the
+ // tight bold-wrapped pattern first, then fall back to a tag-agnostic
+ // boundary match.
  const outlookHeaderRe =
  /<(b|strong)[^>]*>\s*From:?\s*<\/\1>[\s\S]{0,1000}<(b|strong)[^>]*>\s*Sent:?\s*<\/\2>[\s\S]{0,1000}<(b|strong)[^>]*>\s*To:?\s*<\/\3>[\s\S]{0,1000}<(b|strong)[^>]*>\s*Subject:?\s*<\/\4>/i;
  const outlookHeaderMatch = content.match(outlookHeaderRe);
- if (outlookHeaderMatch?.index !== undefined) {
- const fromIdx = outlookHeaderMatch.index;
+ const fromIdx =
+ outlookHeaderMatch?.index ?? findOutlookHeaderTagAgnostic(content);
+ if (fromIdx !== -1) {
  const lookbackStart = Math.max(0, fromIdx - 1000);
  const lookback = content.substring(lookbackStart, fromIdx);
  // Prefer the latest structural divider (border-top div or <hr>)

connectors/gmail/src/gmail-api.ts

@@ -438,12 +438,37 @@ function extractBody(part: GmailMessagePart): { content: string; contentType: "t
  return { content: "", contentType: "text" };
 }
 
+/**
+ * Locates the start of an Outlook-style "From: / Sent: / To: / Subject:"
+ * reply header even when the field labels are not wrapped in `<b>` or
+ * `<strong>` — e.g. corporate Exchange / Outlook variants that put the
+ * label in a `<span style="font-weight:bold">`, a `<font>` tag, or a
+ * plain `MsoNormal` paragraph with no inline bold at all.
+ *
+ * Strategy: replace every HTML tag with a same-length run of spaces so
+ * character offsets in the stripped text still map 1:1 back to the
+ * original. Then require each label to start at a structural boundary
+ * (start of string, a real newline, or 3+ whitespace chars — the smallest
+ * gap any HTML block tag produces when replaced). That anchor is what
+ * keeps user-written prose from false-matching.
+ *
+ * Returns the index of "From:" in the original content, or -1 if no
+ * Outlook reply header is found.
+ */
+export function findOutlookHeaderTagAgnostic(content: string): number {
+ const flat = content.replace(/<[^>]*>/g, (m) => " ".repeat(m.length));
+ const re =
+ /(?<=^|\n|[ \t]{3,})From:[\s\S]{0,1500}?(?<=\n|[ \t]{3,})Sent:[\s\S]{0,800}?(?<=\n|[ \t]{3,})To:[\s\S]{0,1500}?(?<=\n|[ \t]{3,})Subject:/i;
+ const m = flat.match(re);
+ return m?.index ?? -1;
+}
+
 /**
  * Strips quoted reply content from an email body.
  * Since Plot shows each message as a separate note in a thread,
  * the quoted previous messages are redundant noise.
  */
-function stripQuotedReply(
+export function stripQuotedReply(
  content: string,
  contentType: "text" | "html"
 ): string {
@@ -477,16 +502,18 @@ function stripQuotedReply(
  }
 
  // Outlook (desktop, OWA, and corporate Exchange clients) wraps replies
- // with a bolded "From: / Sent: / To: / Subject:" header block. The
- // markup around it varies — sometimes a <div style="border-top:..."
- // divider, sometimes <hr>, sometimes neither — so detect the stable
- // header sequence first, then walk back to the nearest structural
- // boundary.
+ // with a "From: / Sent: / To: / Subject:" header block. The markup
+ // varies — sometimes `<b>` or `<strong>`, sometimes `<span
+ // style="font-weight:bold">`, sometimes a `MsoNormal` paragraph with
+ // no inline bold at all (Gowling-style corporate Exchange). Try the
+ // tight bold-wrapped pattern first, then fall back to a tag-agnostic
+ // boundary match.
  const outlookHeaderRe =
  /<(b|strong)[^>]*>\s*From:?\s*<\/\1>[\s\S]{0,1000}<(b|strong)[^>]*>\s*Sent:?\s*<\/\2>[\s\S]{0,1000}<(b|strong)[^>]*>\s*To:?\s*<\/\3>[\s\S]{0,1000}<(b|strong)[^>]*>\s*Subject:?\s*<\/\4>/i;
  const outlookHeaderMatch = content.match(outlookHeaderRe);
- if (outlookHeaderMatch?.index !== undefined) {
- const fromIdx = outlookHeaderMatch.index;
+ const fromIdx =
+ outlookHeaderMatch?.index ?? findOutlookHeaderTagAgnostic(content);
+ if (fromIdx !== -1) {
  const lookbackStart = Math.max(0, fromIdx - 1000);
  const lookback = content.substring(lookbackStart, fromIdx);
  // Prefer the latest structural divider (border-top div or <hr>)

connectors/gmail/src/gmail-api.ts

@@ -438,12 +438,37 @@ function extractBody(part: GmailMessagePart): { content: string; contentType: "t
  return { content: "", contentType: "text" };
 }
 
+/**
+ * Locates the start of an Outlook-style "From: / Sent: / To: / Subject:"
+ * reply header even when the field labels are not wrapped in `<b>` or
+ * `<strong>` — e.g. corporate Exchange / Outlook variants that put the
+ * label in a `<span style="font-weight:bold">`, a `<font>` tag, or a
+ * plain `MsoNormal` paragraph with no inline bold at all.
+ *
+ * Strategy: replace every HTML tag with a same-length run of spaces so
+ * character offsets in the stripped text still map 1:1 back to the
+ * original. Then require each label to start at a structural boundary
+ * (start of string, a real newline, or 3+ whitespace chars — the smallest
+ * gap any HTML block tag produces when replaced). That anchor is what
+ * keeps user-written prose from false-matching.
+ *
+ * Returns the index of "From:" in the original content, or -1 if no
+ * Outlook reply header is found.
+ */
+export function findOutlookHeaderTagAgnostic(content: string): number {
+ const flat = content.replace(/<[^>]*>/g, (m) => " ".repeat(m.length));
+ const re =
+ /(?<=^|\n|[ \t]{3,})From:[\s\S]{0,1500}?(?<=\n|[ \t]{3,})Sent:[\s\S]{0,800}?(?<=\n|[ \t]{3,})To:[\s\S]{0,1500}?(?<=\n|[ \t]{3,})Subject:/i;
+ const m = flat.match(re);
+ return m?.index ?? -1;
+}
+
 /**
  * Strips quoted reply content from an email body.
  * Since Plot shows each message as a separate note in a thread,
  * the quoted previous messages are redundant noise.
  */
-function stripQuotedReply(
+export function stripQuotedReply(
  content: string,
  contentType: "text" | "html"
 ): string {
@@ -477,16 +502,18 @@ function stripQuotedReply(
  }
 
  // Outlook (desktop, OWA, and corporate Exchange clients) wraps replies
- // with a bolded "From: / Sent: / To: / Subject:" header block. The
- // markup around it varies — sometimes a <div style="border-top:..."
- // divider, sometimes <hr>, sometimes neither — so detect the stable
- // header sequence first, then walk back to the nearest structural
- // boundary.
+ // with a "From: / Sent: / To: / Subject:" header block. The markup
+ // varies — sometimes `<b>` or `<strong>`, sometimes `<span
+ // style="font-weight:bold">`, sometimes a `MsoNormal` paragraph with
+ // no inline bold at all (Gowling-style corporate Exchange). Try the
+ // tight bold-wrapped pattern first, then fall back to a tag-agnostic
+ // boundary match.
  const outlookHeaderRe =
  /<(b|strong)[^>]*>\s*From:?\s*<\/\1>[\s\S]{0,1000}<(b|strong)[^>]*>\s*Sent:?\s*<\/\2>[\s\S]{0,1000}<(b|strong)[^>]*>\s*To:?\s*<\/\3>[\s\S]{0,1000}<(b|strong)[^>]*>\s*Subject:?\s*<\/\4>/i;
  const outlookHeaderMatch = content.match(outlookHeaderRe);
- if (outlookHeaderMatch?.index !== undefined) {
- const fromIdx = outlookHeaderMatch.index;
+ const fromIdx =
+ outlookHeaderMatch?.index ?? findOutlookHeaderTagAgnostic(content);
+ if (fromIdx !== -1) {
  const lookbackStart = Math.max(0, fromIdx - 1000);
  const lookback = content.substring(lookbackStart, fromIdx);
  // Prefer the latest structural divider (border-top div or <hr>)

connectors/gmail/src/gmail-api.ts

@@ -438,12 +438,37 @@ function extractBody(part: GmailMessagePart): { content: string; contentType: "t
  return { content: "", contentType: "text" };
 }
 
+/**
+ * Locates the start of an Outlook-style "From: / Sent: / To: / Subject:"
+ * reply header even when the field labels are not wrapped in `<b>` or
+ * `<strong>` — e.g. corporate Exchange / Outlook variants that put the
+ * label in a `<span style="font-weight:bold">`, a `<font>` tag, or a
+ * plain `MsoNormal` paragraph with no inline bold at all.
+ *
+ * Strategy: replace every HTML tag with a same-length run of spaces so
+ * character offsets in the stripped text still map 1:1 back to the
+ * original. Then require each label to start at a structural boundary
+ * (start of string, a real newline, or 3+ whitespace chars — the smallest
+ * gap any HTML block tag produces when replaced). That anchor is what
+ * keeps user-written prose from false-matching.
+ *
+ * Returns the index of "From:" in the original content, or -1 if no
+ * Outlook reply header is found.
+ */
+export function findOutlookHeaderTagAgnostic(content: string): number {
+ const flat = content.replace(/<[^>]*>/g, (m) => " ".repeat(m.length));
+ const re =
+ /(?<=^|\n|[ \t]{3,})From:[\s\S]{0,1500}?(?<=\n|[ \t]{3,})Sent:[\s\S]{0,800}?(?<=\n|[ \t]{3,})To:[\s\S]{0,1500}?(?<=\n|[ \t]{3,})Subject:/i;
+ const m = flat.match(re);
+ return m?.index ?? -1;
+}
+
 /**
  * Strips quoted reply content from an email body.
  * Since Plot shows each message as a separate note in a thread,
  * the quoted previous messages are redundant noise.
  */
-function stripQuotedReply(
+export function stripQuotedReply(
  content: string,
  contentType: "text" | "html"
 ): string {
@@ -477,16 +502,18 @@ function stripQuotedReply(
  }
 
  // Outlook (desktop, OWA, and corporate Exchange clients) wraps replies
- // with a bolded "From: / Sent: / To: / Subject:" header block. The
- // markup around it varies — sometimes a <div style="border-top:..."
- // divider, sometimes <hr>, sometimes neither — so detect the stable
- // header sequence first, then walk back to the nearest structural
- // boundary.
+ // with a "From: / Sent: / To: / Subject:" header block. The markup
+ // varies — sometimes `<b>` or `<strong>`, sometimes `<span
+ // style="font-weight:bold">`, sometimes a `MsoNormal` paragraph with
+ // no inline bold at all (Gowling-style corporate Exchange). Try the
+ // tight bold-wrapped pattern first, then fall back to a tag-agnostic
+ // boundary match.
  const outlookHeaderRe =
  /<(b|strong)[^>]*>\s*From:?\s*<\/\1>[\s\S]{0,1000}<(b|strong)[^>]*>\s*Sent:?\s*<\/\2>[\s\S]{0,1000}<(b|strong)[^>]*>\s*To:?\s*<\/\3>[\s\S]{0,1000}<(b|strong)[^>]*>\s*Subject:?\s*<\/\4>/i;
  const outlookHeaderMatch = content.match(outlookHeaderRe);
- if (outlookHeaderMatch?.index !== undefined) {
- const fromIdx = outlookHeaderMatch.index;
+ const fromIdx =
+ outlookHeaderMatch?.index ?? findOutlookHeaderTagAgnostic(content);
+ if (fromIdx !== -1) {
  const lookbackStart = Math.max(0, fromIdx - 1000);
  const lookback = content.substring(lookbackStart, fromIdx);
  // Prefer the latest structural divider (border-top div or <hr>)