test(issues): document Greptile mid-token case vs old strip behavior

Made-with: Cursor

server/src/__tests__/normalize-agent-mention-token.test.ts

@@ -14,8 +14,12 @@ describe("normalizeAgentMentionToken", () => {
     expect(normalizeAgentMentionToken("Baba ")).toBe("Baba");
   });
 
-  // Greptile: entity mid-token (not only trailing) — must decode & to &, not delete the sequence.
-  it("decodes & in the middle of a mention token", () => {
+  /**
+   * Greptile (PR review): cover an entity in the *middle* of a token, not only trailing ` ` / ` `.
+   * Earlier implementation *stripped* `&`, turning `Ba&ba` into `Baba` and breaking names like `M&M`.
+   * Current behavior decodes `&` → `&`, so the expected value is `Ba&ba`, not `Baba`.
+   */
+  it("decodes a named entity mid-token (Ba&ba → Ba&ba, not strip to Baba)", () => {
     expect(normalizeAgentMentionToken("Ba&ba")).toBe("Ba&ba");
   });