diff --git a/CHANGELOG.md b/CHANGELOG.md
index 1c3e8767dc..a245b7ccee 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -13,6 +13,7 @@ Full release notes with details on each version: [GitHub Releases](https://githu
 - Fix: the incremental rebuild no longer purges AST nodes it just reported as fail-closed "kept" — the eviction pass re-checks the kept set, so a moved-file/symlink layout can't deadlock the shrink guard into refusing every update (#3697, #3695, thanks @hopstreax).
 - Fix: `graph.html` no longer crashes vis-network with a stack overflow on large graphs — nodes are seeded on a spiral before physics runs so overlap-avoidance can't blow the layout recursion (#3699, thanks @sanjaiyan-dev).
 - Fix: node and edge tooltips now show special characters literally (C++ templates like `vector<int>`, generics, `&`, quotes) instead of raw HTML entities, while the HTML sinks that need escaping keep it (#3686, #3664, thanks @hopstreax).
+- Fix: `graphify install` no longer rewrites an existing `CLAUDE.md`/`GEMINI.md` to CRLF on Windows just to append its registration block, and the re-install guard now matches the actual heading it writes instead of a bare `"graphify"` substring check — a file that merely mentions the project elsewhere is no longer treated as already registered, and a stale or hand-edited block is now refreshed on re-install rather than silently left alone (#3668, thanks @kevinishii-spec).
 - Docs: repository links now point at `Graphify-Labs/graphify` instead of the old account (including in generated wiki output), translated READMEs use the current logo, GitHub issue/PR templates were added, and the Enterprise link was corrected (#3692, #3694, #3693, thanks @Abdul535).
 
 ## 0.9.64 (2026-09-18)
diff --git a/graphify/install.py b/graphify/install.py
index 2fb2192760..0001daa4dd 100644
--- a/graphify/install.py
+++ b/graphify/install.py
@@ -299,11 +299,16 @@ def _remove_claude_skill_registration(project_dir: Path) -> None:
     content = claude_md.read_text(encoding="utf-8")
     # Match the exact H1 `# graphify` registration heading, never a substring of a
     # user's `## graphify`/`### graphify` (#2062). Section runs to the next H1.
-    cleaned = _remove_marker_section(content, "# graphify", boundary_prefix="# ")
+    # _SKILL_REGISTRATION_MARKER is the single source of truth for this heading,
+    # shared with _register_always_on_block so an insert and its removal always
+    # agree on what they're matching (#3668).
+    cleaned = _remove_marker_section(content, _SKILL_REGISTRATION_MARKER, boundary_prefix="# ")
     if cleaned is None:
         return
     if cleaned:
-        claude_md.write_text(cleaned + "\n", encoding="utf-8")
+        # newline="" so the rest of the file's own line endings are never
+        # translated on write (#3668, same CRLF issue as the insert side).
+        claude_md.write_text(cleaned + "\n", encoding="utf-8", newline="")
         print(f"  CLAUDE.md        ->  graphify skill registration removed from {claude_md}")
     else:
         claude_md.unlink()
@@ -358,15 +363,30 @@ def _claude_pretooluse_hooks(strict: bool = False, project: bool = False) -> "li
          "hooks": [{"type": "command", "command": read_cmd, "timeout": 10}]},
     ]
 def _skill_registration(skill_path: str = "~/.claude/skills/graphify/SKILL.md") -> str:
+    # Heading is "# graphify" (H1) to match _SKILL_REGISTRATION_MARKER, which
+    # _register_always_on_block anchors its idempotent replace-or-append on.
     return (
-        "\n# graphify\n"
+        "# graphify\n"
         f"- **graphify** (`{skill_path}`) "
         "- any input to knowledge graph. Trigger: `/graphify`\n"
         "When the user types `/graphify`, use the installed graphify skill "
         "or instructions before doing anything else.\n"
     )
 def _register_always_on_block(target: Path, prefix: str, registration: str) -> None:
-    """Append an always-on registration to *target*, degrading instead of raising.
+    """Idempotently add or refresh an always-on registration in *target*, degrading
+    instead of raising.
+
+    Uses _replace_or_append_section (the same marker-anchored helper claude_install
+    and gemini_install already use, here with the H1 boundary_prefix so it stays
+    paired with _remove_claude_skill_registration's own H1 match) rather than a
+    bare append, so a stale or edited block gets refreshed on re-install instead
+    of the bare "graphify" substring check silently treating any unrelated
+    mention of the word as already-registered and skipping every re-run (#3668).
+
+    Written with newline="" so an existing file's own line endings are never
+    translated -- Path.write_text otherwise opens in text mode, which on Windows
+    turns the WHOLE file's pre-existing bare-LF content into CRLF just to append a
+    few lines (#3668).
 
     The skill files are copied before this runs, so a *target* that cannot be
     read or written must not abort an otherwise-complete install (#3474). That
@@ -375,16 +395,19 @@ def _register_always_on_block(target: Path, prefix: str, registration: str) -> N
     stow with read-only sources leave the same shape.
     """
     try:
-        if target.exists():
-            content = target.read_text(encoding="utf-8")
-            if "graphify" in content:
-                print(f"{prefix}already registered (no change)")
-            else:
-                target.write_text(content.rstrip() + registration, encoding="utf-8")
-                print(f"{prefix}skill registered in {target}")
+        existed = target.exists()
+        content = target.read_text(encoding="utf-8") if existed else ""
+        new_content = _replace_or_append_section(
+            content, _SKILL_REGISTRATION_MARKER, registration, boundary_prefix="# "
+        )
+        if existed and new_content == content:
+            print(f"{prefix}already registered (no change)")
+        elif existed:
+            target.write_text(new_content, encoding="utf-8", newline="")
+            print(f"{prefix}skill registered in {target}")
         else:
             target.parent.mkdir(parents=True, exist_ok=True)
-            target.write_text(registration.lstrip(), encoding="utf-8")
+            target.write_text(new_content, encoding="utf-8", newline="")
             print(f"{prefix}created at {target}")
     except OSError as exc:
         print(f"{prefix}skipped: {exc.__class__.__name__}: {exc}", file=sys.stderr)
@@ -542,7 +565,9 @@ def _register_always_on_block(target: Path, prefix: str, registration: str) -> N
 def _canonical_platform(platform_name: str) -> str:
     """Resolve a CLI platform alias to its real _PLATFORM_CONFIG key."""
     return _PLATFORM_ALIASES.get(platform_name, platform_name)
-def _replace_or_append_section(content: str, marker: str, new_section: str) -> str:
+def _replace_or_append_section(
+    content: str, marker: str, new_section: str, boundary_prefix: str = "## "
+) -> str:
     """Idempotently update or append a graphify-owned section in shared files.
 
     If no line is exactly ``marker`` (the heading, at column 0), append
@@ -550,10 +575,15 @@ def _replace_or_append_section(content: str, marker: str, new_section: str) -> s
     content).
 
     If a real ``marker`` heading exists, replace the existing section in place.
-    The section runs from that heading to the line before the next H2 heading
-    (``## `` at line start), or to EOF if no later H2 exists. This lets older
-    installs receive the updated copy without users having to uninstall and
-    reinstall (issue #580).
+    The section runs from that heading to the line before the next
+    ``boundary_prefix`` heading (default the next H2), or to EOF if none
+    follows. This lets older installs receive the updated copy without users
+    having to uninstall and reinstall (issue #580).
+
+    ``boundary_prefix`` must match whatever level ``marker`` itself is (``"# "``
+    for an H1 marker, the default ``"## "`` for an H2 one) — mirrors
+    ``_remove_marker_section``'s own ``boundary_prefix`` so an insert and its
+    matching removal agree on where a section ends (#3668).
 
     The heading is matched only when a line *is* exactly ``marker`` (after
     stripping surrounding whitespace), never as a substring. Matching ``##
@@ -572,7 +602,7 @@ def _replace_or_append_section(content: str, marker: str, new_section: str) -> s
     start = starts[-1]
     end = len(lines)
     for j in range(start + 1, len(lines)):
-        if lines[j].startswith("## "):
+        if lines[j].startswith(boundary_prefix):
             end = j
             break
 
@@ -746,6 +776,14 @@ def _print_install_usage() -> None:
 _CODEBUDDY_MD_MARKER = "## graphify"
 _AGENTS_MD_MARKER = "## graphify"
 _GEMINI_MD_MARKER = "## graphify"
+# Deliberately H1, not H2 like the markers above: this one anchors the SKILL
+# registration block _register_always_on_block writes into .claude/CLAUDE.md
+# (or CODEBUDDY.md), a different section from the "always-on instructions"
+# block the H2 markers above anchor. Kept at H1 specifically so it can never
+# collide with a genuine user-authored "## graphify" heading elsewhere in the
+# same file (#2062) — _remove_claude_skill_registration matches on this same
+# constant so the two stay in sync.
+_SKILL_REGISTRATION_MARKER = "# graphify"
 def _gemini_hook(project: bool = False) -> dict:
     """Gemini CLI BeforeTool hook, resolved to a shell-agnostic `graphify` call.
 
@@ -777,7 +815,10 @@ def gemini_install(project_dir: Path | None = None, *, project: bool = False) ->
     if target.exists() and new_content == target.read_text(encoding="utf-8"):
         print(f"graphify already configured in {target.resolve()} (no change)")
     else:
-        target.write_text(new_content, encoding="utf-8")
+        # newline="" so an existing file's own line endings are never translated
+        # (Path.write_text otherwise opens in text mode, which on Windows turns
+        # the WHOLE file's pre-existing bare-LF content into CRLF, #3668).
+        target.write_text(new_content, encoding="utf-8", newline="")
         print(f"graphify section written to {target.resolve()}")
 
     # Always re-install the Gemini hook so an older payload (e.g. pre-issue-#580
diff --git a/tests/test_install.py b/tests/test_install.py
index 72d4e4dd26..8caa61a6c4 100644
--- a/tests/test_install.py
+++ b/tests/test_install.py
@@ -199,6 +199,91 @@ def test_install_claude_md_success_output_unchanged(tmp_path, monkeypatch, capsy
     assert "  CLAUDE.md        ->  already registered (no change)" in second
 
 
+def test_install_claude_md_does_not_skip_on_an_unrelated_mention_of_the_word(tmp_path, monkeypatch):
+    """#3668: the idempotency guard used to be a bare `"graphify" in content`
+    substring check, so any pre-existing mention of the word anywhere in the
+    file (a note to self, an unrelated project instruction) was wrongly
+    treated as "already registered" and the real block never got written."""
+    from graphify.__main__ import install
+
+    home = tmp_path / "home"
+    home.mkdir()
+    claude_md = home / ".claude" / "CLAUDE.md"
+    claude_md.parent.mkdir(parents=True)
+    claude_md.write_text("See https://github.com/Graphify-Labs/graphify for details.\n", encoding="utf-8")
+
+    monkeypatch.delenv("CLAUDE_CONFIG_DIR", raising=False)
+    monkeypatch.chdir(tmp_path)
+    with patch("graphify.__main__.Path.home", return_value=home):
+        install(platform="claude")
+
+    content = claude_md.read_text(encoding="utf-8")
+    assert "# graphify\n" in content, (
+        f"an unrelated mention of the word must not suppress the real "
+        f"registration block; got {content!r}"
+    )
+    assert "See https://github.com/Graphify-Labs/graphify for details." in content, (
+        "the user's own pre-existing content must survive"
+    )
+
+
+def test_install_claude_md_refreshes_a_stale_registration_block(tmp_path, monkeypatch):
+    """#3668: a previously-installed block that has since been hand-edited (or
+    predates a skill-path change) must be refreshed on re-install, not
+    silently left stale because the bare word "graphify" is still present."""
+    from graphify.__main__ import install
+
+    home = tmp_path / "home"
+    home.mkdir()
+    claude_md = home / ".claude" / "CLAUDE.md"
+    claude_md.parent.mkdir(parents=True)
+    claude_md.write_text(
+        "# graphify\n- an old, hand-edited line that does not match the "
+        "current registration text\n",
+        encoding="utf-8",
+    )
+
+    monkeypatch.delenv("CLAUDE_CONFIG_DIR", raising=False)
+    monkeypatch.chdir(tmp_path)
+    with patch("graphify.__main__.Path.home", return_value=home):
+        install(platform="claude")
+
+    content = claude_md.read_text(encoding="utf-8")
+    assert "hand-edited line" not in content, "the stale block must be replaced, not kept"
+    assert "Trigger: `/graphify`" in content, "the current registration text must be written"
+
+
+def test_register_always_on_block_writes_without_newline_translation(tmp_path, monkeypatch):
+    """#3668: Path.write_text opens in text mode, which on Windows turns a
+    pre-existing bare-LF file's WHOLE content into CRLF just to append a few
+    lines. newline="" must be passed so no translation happens. The bug
+    itself is only observable on Windows, so this checks the call was made
+    correctly rather than depending on the host OS's own newline handling."""
+    from graphify import install as install_mod
+
+    target = tmp_path / "CLAUDE.md"
+    target.write_text("Some existing notes.\n", encoding="utf-8")
+
+    calls: list[dict] = []
+    orig_write_text = Path.write_text
+
+    def _tracking_write_text(self, *args, **kwargs):
+        calls.append(kwargs)
+        return orig_write_text(self, *args, **kwargs)
+
+    monkeypatch.setattr(Path, "write_text", _tracking_write_text)
+
+    install_mod._register_always_on_block(
+        target, "  CLAUDE.md        ->  ", install_mod._skill_registration()
+    )
+
+    assert calls, "write_text should have been called"
+    assert calls[-1].get("newline") == "", (
+        f"write_text must pass newline='' so the rest of the file's line "
+        f"endings are never translated; got kwargs {calls[-1]!r}"
+    )
+
+
 def test_install_codebuddy(tmp_path):
     _install(tmp_path, "codebuddy")
     assert (tmp_path / ".codebuddy" / "skills" / "graphify" / "SKILL.md").exists()
@@ -1130,6 +1215,36 @@ def test_gemini_install_merges_existing_gemini_md(tmp_path):
     assert "graphify-out/GRAPH_REPORT.md" in content
 
 
+def test_gemini_install_writes_gemini_md_without_newline_translation(tmp_path, monkeypatch):
+    """#3668: same CRLF issue as _register_always_on_block, here in the
+    GEMINI.md write. Path.write_text opens in text mode, which on Windows
+    would turn a pre-existing bare-LF GEMINI.md's WHOLE content into CRLF
+    just to merge in a few lines. Checks the call was made correctly rather
+    than depending on the host OS's own newline handling."""
+    from graphify.__main__ import gemini_install
+
+    gemini_md = tmp_path / "GEMINI.md"
+    gemini_md.write_text("# My project rules\n", encoding="utf-8")
+
+    calls: list[dict] = []
+    orig_write_text = Path.write_text
+
+    def _tracking_write_text(self, *args, **kwargs):
+        if self == gemini_md:
+            calls.append(kwargs)
+        return orig_write_text(self, *args, **kwargs)
+
+    monkeypatch.setattr(Path, "write_text", _tracking_write_text)
+
+    gemini_install(tmp_path)
+
+    assert calls, "write_text should have been called for GEMINI.md"
+    assert calls[-1].get("newline") == "", (
+        f"write_text must pass newline='' so the rest of the file's line "
+        f"endings are never translated; got kwargs {calls[-1]!r}"
+    )
+
+
 def test_gemini_uninstall_removes_section(tmp_path):
     from graphify.__main__ import gemini_install, gemini_uninstall