fix(slack): Harden mrkdwn normalization edges

Only unwrap raw URLs when formatting encloses a standalone URL token. This avoids corrupting emphasized prose that merely starts with a link.

Strip full HTML comments until stable and escape any leftover opener fragments so comment removal cannot reintroduce a literal comment start.

Co-Authored-By: Codex GPT-5 <noreply@openai.com>

Author: dcramer

packages/junior/src/chat/slack/mrkdwn.ts

@@ -48,8 +48,26 @@ function normalizeCommonMarkEmphasis(text: string): string {
     .replace(/~~([^\s~](?:[^\n]*?[^\s~])?)~~/g, "~$1~");
 }
 
+function removeMarkdownHtmlComments(text: string): string {
+  let normalized = text;
+
+  while (true) {
+    const start = normalized.indexOf("<!--");
+    if (start === -1) {
+      return normalized;
+    }
+
+    const end = normalized.indexOf("-->", start + 4);
+    if (end === -1) {
+      return normalized;
+    }
+
+    normalized = `${normalized.slice(0, start)}${normalized.slice(end + 3)}`;
+  }
+}
+
 function stripMarkdownHtmlComments(text: string): string {
-  return text.replace(/<!--[\s\S]*?-->/g, "");
+  return removeMarkdownHtmlComments(text).replaceAll("<!--", "&lt;!--");
 }
 
 function escapeSlackLinkLabel(text: string): string {
@@ -317,18 +335,51 @@ function splitSlackUrlSuffix(text: string): { suffix: string; url: string } {
   return { suffix, url };
 }
 
+function splitWrappedRawUrlToken(token: string): {
+  after: string;
+  before: string;
+  core: string;
+} {
+  let start = 0;
+  while (/[([{'"']/.test(token[start] ?? "")) {
+    start += 1;
+  }
+
+  let end = token.length;
+  while (end > start && /[)\]}"',.!?;:]/.test(token[end - 1] ?? "")) {
+    end -= 1;
+  }
+
+  return {
+    after: token.slice(end),
+    before: token.slice(0, start),
+    core: token.slice(start, end),
+  };
+}
+
+function normalizeWrappedRawUrlToken(token: string): string {
+  const { after, before, core } = splitWrappedRawUrlToken(token);
+  const match = core.match(/^([*_~`]+)(https?:\/\/[^\s<]+?)([*_~`]+)$/);
+  if (!match) {
+    return token;
+  }
+
+  const { suffix, url } = splitSlackUrlSuffix(match[2] ?? "");
+  return `${before}${formatSlackLink(url)}${suffix}${after}`;
+}
+
 function normalizeWrappedRawUrls(text: string): string {
-  return text.replace(
-    /([*_~`]+)?(https?:\/\/[^\s<]+)([*_~`]+)?/g,
-    (match, leading, rawUrl, trailing) => {
-      if (!leading && !trailing) {
-        return match;
-      }
+  let out = "";
+  let lastIndex = 0;
 
-      const { suffix, url } = splitSlackUrlSuffix(String(rawUrl));
-      return `${formatSlackLink(url)}${suffix}`;
-    },
-  );
+  for (const match of text.matchAll(/\S+/g)) {
+    const index = match.index ?? 0;
+    out += text.slice(lastIndex, index);
+    out += normalizeWrappedRawUrlToken(match[0]);
+    lastIndex = index + match[0].length;
+  }
+
+  return `${out}${text.slice(lastIndex)}`;
 }
 
 function parseMarkdownTableRow(line: string): string[] | null {

packages/junior/tests/unit/misc/output.test.ts

@@ -96,12 +96,24 @@ describe("renderSlackMrkdwn", () => {
     );
   });
 
+  it("leaves raw urls inside larger formatted prose untouched", () => {
+    expect(renderSlackMrkdwn("*https://docs.slack.dev/ more context*")).toBe(
+      "*https://docs.slack.dev/ more context*",
+    );
+  });
+
   it("strips html comments outside code fences", () => {
     expect(
       renderSlackMrkdwn("Intro\n<!-- internal note -->\n## Summary\nDone"),
     ).toBe("Intro\n\n*Summary*\n\nDone");
   });
 
+  it("escapes dangling html comment openers after stripping comments", () => {
+    expect(renderSlackMrkdwn("Intro <!<!-- internal note -->-- outro")).toBe(
+      "Intro &lt;!-- outro",
+    );
+  });
+
   it("preserves already-valid Slack links, mentions, and formatting", () => {
     expect(
       renderSlackMrkdwn("<https://docs.slack.dev/|docs> <@U123> *ready*"),