diff --git a/src/utils/file-system.ts b/src/utils/file-system.ts
index 7306015..de1b26e 100644
--- a/src/utils/file-system.ts
+++ b/src/utils/file-system.ts
@@ -91,7 +91,16 @@ export function parseGitignoreLines(content: string): Set<string> {
 
 /**
  * Replace or remove a markdown section identified by a heading marker.
- * Finds the marker, then locates the next level-2 heading to determine section bounds.
+ *
+ * The section runs from its marker heading to the next heading of the same or
+ * higher level (a level-1 `#` or level-2 `##` heading). Deeper headings (`###`
+ * and below) are treated as part of the section body.
+ *
+ * Bounding by the next same-or-higher-level heading matters for correctness:
+ * the previous implementation only looked for the next `## ` heading, so a
+ * trailing top-level `# heading` — or any user content that followed the
+ * managed section — was swallowed into it and deleted when the section was
+ * removed (e.g. during `bmalph reset`).
  *
  * @param content - The full file content
  * @param marker - The section heading to find (e.g. "## BMAD-METHOD Integration")
@@ -105,10 +114,16 @@ export function replaceSection(content: string, marker: string, replacement: str
   const before = content.slice(0, sectionStart);
   const afterSection = content.slice(sectionStart);
 
-  const headingText = marker.startsWith("## ") ? marker.slice(3) : marker;
-  const headingTextEscaped = headingText.replace(/[.*+?^${}()|[\]\\]/g, "\\$&");
-  const nextHeadingMatch = afterSection.match(new RegExp(`\\n## (?!${headingTextEscaped})`));
-  const after = nextHeadingMatch ? afterSection.slice(nextHeadingMatch.index ?? 0) : "";
+  // Skip past the marker's own heading line so it is never matched as the
+  // section boundary, then find the next level-1 or level-2 heading.
+  const markerLineBreak = afterSection.indexOf("\n");
+  const afterMarkerLine = markerLineBreak === -1 ? "" : afterSection.slice(markerLineBreak);
+
+  const nextHeadingMatch = afterMarkerLine.match(/\n#{1,2} /);
+  const after =
+    nextHeadingMatch && nextHeadingMatch.index !== undefined
+      ? afterMarkerLine.slice(nextHeadingMatch.index)
+      : "";
 
   return before.trimEnd() + replacement + after;
 }
diff --git a/tests/utils/file-system.test.ts b/tests/utils/file-system.test.ts
index 23f3e87..70b52cf 100644
--- a/tests/utils/file-system.test.ts
+++ b/tests/utils/file-system.test.ts
@@ -354,5 +354,57 @@ describe("isDirectory() with mocked fs", () => {
       expect(result).toContain("Replaced.");
       expect(result).not.toContain("Last content.");
     });
+
+    it("preserves a trailing top-level heading when removing the section", () => {
+      const doc = "# Project\n\n## BMAD\n\nManaged body.\n\n# My Notes\n\nKeep this.";
+      const result = replaceSection(doc, "## BMAD", "");
+
+      expect(result).not.toContain("Managed body.");
+      expect(result).toContain("# My Notes");
+      expect(result).toContain("Keep this.");
+    });
+
+    it("treats nested (### and deeper) headings as part of the section body", () => {
+      const doc = "## BMAD\n\nIntro.\n\n### Sub A\n\nDetails.\n\n## After\n\nUser section.";
+      const result = replaceSection(doc, "## BMAD", "");
+
+      // Everything in the BMAD section, including its ### subheading, is removed...
+      expect(result).not.toContain("Intro.");
+      expect(result).not.toContain("### Sub A");
+      expect(result).not.toContain("Details.");
+      // ...but the following same-level section survives.
+      expect(result).toContain("## After");
+      expect(result).toContain("User section.");
+    });
+
+    it("does not match the marker's own heading as the section boundary", () => {
+      const doc = "## BMAD\n\nold body\n\n## Other\n\nother body";
+      const result = replaceSection(doc, "## BMAD", "\n## BMAD\n\nnew body\n");
+
+      expect(result).toContain("new body");
+      expect(result).not.toContain("old body");
+      expect(result).toContain("## Other");
+      expect(result).toContain("other body");
+    });
+
+    it("preserves a trailing heading and its content on CRLF (Windows) files", () => {
+      const doc = "# Project\r\n\r\n## BMAD\r\n\r\nManaged body.\r\n\r\n# Notes\r\n\r\nKeep this.";
+      const result = replaceSection(doc, "## BMAD", "");
+
+      // The managed body is removed, but the user's trailing section survives.
+      expect(result).not.toContain("Managed body.");
+      expect(result).toContain("# Notes");
+      expect(result).toContain("Keep this.");
+    });
+
+    it("documents the limitation: headerless trailing prose is absorbed into the section", () => {
+      // Known limitation — without a heading boundary there is nothing to
+      // distinguish trailing prose from the managed section body, so it is
+      // removed. Pinned here so the behavior change is intentional, not silent.
+      const doc = "## BMAD\n\nManaged body.\n\nTrailing prose with no heading.";
+      const result = replaceSection(doc, "## BMAD", "");
+
+      expect(result).not.toContain("Trailing prose with no heading.");
+    });
   });
 });