diff --git a/cli.py b/cli.py
index 3b1ecd8ae42f..7b3c75ce10db 100644
--- a/cli.py
+++ b/cli.py
@@ -66,6 +66,10 @@
     format_token_count_compact,
 )
 from hermes_cli.banner import _format_context_length, format_banner_version_label
+from hermes_cli.paste_collapse import (
+    expand_paste_references,
+    materialize_paste_for_insertion,
+)
 
 _COMMAND_SPINNER_FRAMES = ("⠋", "⠙", "⠹", "⠸", "⠼", "⠴", "⠦", "⠧", "⠇", "⠏")
 
@@ -9171,26 +9175,25 @@ def handle_paste(event):
             pasted_text = pasted_text.replace('\r\n', '\n').replace('\r', '\n')
             if _should_auto_attach_clipboard_image_on_paste(pasted_text) and self._try_attach_clipboard_image():
                 event.app.invalidate()
-            if pasted_text:
-                # Sanitize surrogate characters (e.g. from Word/Google Docs paste) before writing
-                from run_agent import _sanitize_surrogates
-                pasted_text = _sanitize_surrogates(pasted_text)
-                line_count = pasted_text.count('\n')
-                buf = event.current_buffer
-                if line_count >= 5 and not buf.text.strip().startswith('/'):
-                    _paste_counter[0] += 1
-                    paste_dir = _hermes_home / "pastes"
-                    paste_dir.mkdir(parents=True, exist_ok=True)
-                    paste_file = paste_dir / f"paste_{_paste_counter[0]}_{datetime.now().strftime('%H%M%S')}.txt"
-                    paste_file.write_text(pasted_text, encoding="utf-8")
-                    placeholder = f"[Pasted text #{_paste_counter[0]}: {line_count + 1} lines \u2192 {paste_file}]"
-                    prefix = ""
-                    if buf.cursor_position > 0 and buf.text[buf.cursor_position - 1] != '\n':
-                        prefix = "\n"
-                    _paste_just_collapsed[0] = True
-                    buf.insert_text(prefix + placeholder)
-                else:
-                    buf.insert_text(pasted_text)
+            if not pasted_text:
+                return
+
+            # Sanitize surrogate characters (e.g. from Word/Google Docs paste) before writing
+            from run_agent import _sanitize_surrogates
+            pasted_text = _sanitize_surrogates(pasted_text)
+
+            next_counter = _paste_counter[0] + 1
+            inserted_text, collapsed = materialize_paste_for_insertion(
+                pasted_text,
+                current_buffer_text=event.current_buffer.text or "",
+                paste_dir=_hermes_home / "pastes",
+                counter=next_counter,
+                now=datetime.now(),
+            )
+            if collapsed:
+                _paste_counter[0] = next_counter
+                _paste_just_collapsed[0] = True
+            event.current_buffer.insert_text(inserted_text)
 
         @kb.add('c-v')
         def handle_ctrl_v(event):
@@ -9283,7 +9286,9 @@ def _input_height():
 
         input_area.window.height = _input_height
 
-        # Paste collapsing: detect large pastes and save to temp file
+        # Paste collapsing: large bracketed pastes are collapsed to file references.
+        # The decision now happens in the paste handler so we only ever collapse
+        # the pasted chunk, not the whole draft buffer.
         _paste_counter = [0]
         _prev_text_len = [0]
         _prev_newline_count = [0]
@@ -10039,40 +10044,29 @@ def process_loop():
                                 app.exit()
                         continue
                     
-                    # Expand paste references back to full content
-                    import re as _re
-                    _paste_ref_re = _re.compile(r'\[Pasted text #\d+: \d+ lines \u2192 (.+?)\]')
-                    paste_refs = list(_paste_ref_re.finditer(user_input)) if isinstance(user_input, str) else []
-                    if paste_refs:
-                        def _expand_ref(m):
-                            p = Path(m.group(1))
-                            return p.read_text(encoding="utf-8") if p.exists() else m.group(0)
-                        expanded = _paste_ref_re.sub(_expand_ref, user_input)
-                        total_lines = expanded.count('\n') + 1
-                        n_pastes = len(paste_refs)
-                        _user_bar = f"[{_accent_hex()}]{'─' * 40}[/]"
+                    # Expand paste references back to full content for the
+                    # actual agent payload, but keep the compact placeholders in
+                    # the local echo so large pasted chunks do not flood the TUI.
+                    display_input = user_input
+                    expanded_user_input = (
+                        expand_paste_references(user_input)
+                        if isinstance(user_input, str)
+                        else user_input
+                    )
+
+                    _user_bar = f"[{_accent_hex()}]{'─' * 40}[/]"
+                    if isinstance(display_input, str) and display_input != expanded_user_input:
+                        expanded_line_count = expanded_user_input.count('\n') + 1 if isinstance(expanded_user_input, str) else 1
                         print()
                         ChatConsole().print(_user_bar)
-                        # Show any surrounding user text alongside the paste summary
-                        split_parts = _paste_ref_re.split(user_input)
-                        visible_user_text = " ".join(
-                            split_parts[i].strip() for i in range(0, len(split_parts), 2) if split_parts[i].strip()
+                        ChatConsole().print(
+                            f"[bold {_accent_hex()}]●[/] [bold]{_escape(f'[Pasted text: {expanded_line_count} lines]')}[/]"
                         )
-                        if visible_user_text:
-                            ChatConsole().print(
-                                f"[bold {_accent_hex()}]\u25cf[/] [bold]{_escape(visible_user_text)}[/] "
-                                f"[dim]({n_pastes} pasted block{'s' if n_pastes > 1 else ''}, {total_lines} lines total)[/]"
-                            )
-                        else:
-                            ChatConsole().print(
-                                f"[bold {_accent_hex()}]\u25cf[/] [bold]{_escape(f'[Pasted text: {total_lines} lines]')}[/]"
-                            )
-                        user_input = expanded
                     else:
                         _user_bar = f"[{_accent_hex()}]{'─' * 40}[/]"
-                        if '\n' in user_input:
-                            first_line = user_input.split('\n')[0]
-                            line_count = user_input.count('\n') + 1
+                        if '\n' in display_input:
+                            first_line = display_input.split('\n')[0]
+                            line_count = display_input.count('\n') + 1
                             print()
                             ChatConsole().print(_user_bar)
                             ChatConsole().print(
@@ -10082,7 +10076,7 @@ def _expand_ref(m):
                         else:
                             print()
                             ChatConsole().print(_user_bar)
-                            ChatConsole().print(f"[bold {_accent_hex()}]●[/] [bold]{_escape(user_input)}[/]")
+                            ChatConsole().print(f"[bold {_accent_hex()}]●[/] [bold]{_escape(display_input)}[/]")
                     
                     # Show image attachment count
                     if submit_images:
@@ -10094,7 +10088,7 @@ def _expand_ref(m):
                     app.invalidate()  # Refresh status line
 
                     try:
-                        self.chat(user_input, images=submit_images or None)
+                        self.chat(expanded_user_input, images=submit_images or None)
                     finally:
                         self._agent_running = False
                         self._spinner_text = ""
diff --git a/hermes_cli/paste_collapse.py b/hermes_cli/paste_collapse.py
new file mode 100644
index 000000000000..45cfaa7adebd
--- /dev/null
+++ b/hermes_cli/paste_collapse.py
@@ -0,0 +1,66 @@
+from __future__ import annotations
+
+from datetime import datetime
+from pathlib import Path
+import re
+
+PASTE_REF_RE = re.compile(r"\[Pasted text #(\d+): \d+ lines → (.+?)\]")
+
+
+def should_collapse_pasted_text(pasted_text: str, *, min_lines: int = 5) -> bool:
+    if not pasted_text:
+        return False
+    return pasted_text.count("\n") >= min_lines
+
+
+def write_pasted_text_reference(
+    pasted_text: str,
+    *,
+    paste_dir: Path,
+    counter: int,
+    now: datetime,
+) -> str:
+    paste_dir.mkdir(parents=True, exist_ok=True)
+    paste_file = paste_dir / f"paste_{counter}_{now.strftime('%H%M%S')}.txt"
+    paste_file.write_text(pasted_text, encoding="utf-8")
+    line_count = pasted_text.count("\n") + 1
+    return f"[Pasted text #{counter}: {line_count} lines → {paste_file}]"
+
+
+def materialize_paste_for_insertion(
+    pasted_text: str,
+    *,
+    current_buffer_text: str,
+    paste_dir: Path,
+    counter: int,
+    now: datetime,
+    min_lines: int = 5,
+) -> tuple[str, bool]:
+    if not pasted_text:
+        return "", False
+    if (current_buffer_text or "").startswith("/"):
+        return pasted_text, False
+    if not should_collapse_pasted_text(pasted_text, min_lines=min_lines):
+        return pasted_text, False
+    return (
+        write_pasted_text_reference(
+            pasted_text,
+            paste_dir=paste_dir,
+            counter=counter,
+            now=now,
+        ),
+        True,
+    )
+
+
+def expand_paste_references(text: str) -> str:
+    if not text:
+        return text
+
+    def repl(match: re.Match[str]) -> str:
+        paste_path = Path(match.group(2))
+        if not paste_path.exists():
+            return match.group(0)
+        return paste_path.read_text(encoding="utf-8")
+
+    return PASTE_REF_RE.sub(repl, text)
diff --git a/tests/test_cli_paste_collapse.py b/tests/test_cli_paste_collapse.py
new file mode 100644
index 000000000000..4a2c6f0c1ef9
--- /dev/null
+++ b/tests/test_cli_paste_collapse.py
@@ -0,0 +1,132 @@
+from datetime import datetime
+
+from hermes_cli.paste_collapse import (
+    expand_paste_references,
+    materialize_paste_for_insertion,
+    should_collapse_pasted_text,
+    write_pasted_text_reference,
+)
+
+
+class FakeBuffer:
+    def __init__(self, text: str, cursor_position: int):
+        self.text = text
+        self.cursor_position = cursor_position
+
+    def insert_text(self, value: str):
+        self.text = self.text[: self.cursor_position] + value + self.text[self.cursor_position :]
+        self.cursor_position += len(value)
+
+
+def test_should_not_collapse_small_single_word_paste():
+    assert should_collapse_pasted_text("hello") is False
+
+
+def test_should_not_collapse_short_multiline_paste_under_threshold():
+    assert should_collapse_pasted_text("a\nb\nc\nd") is False
+
+
+def test_should_collapse_large_multiline_paste_at_threshold():
+    pasted = "1\n2\n3\n4\n5\n6"
+    assert should_collapse_pasted_text(pasted) is True
+
+
+def test_write_reference_creates_file_and_returns_placeholder(tmp_path):
+    ref = write_pasted_text_reference(
+        "alpha\nbeta\ngamma\ndelta\nepsilon\nzeta",
+        paste_dir=tmp_path,
+        counter=1,
+        now=datetime(2026, 3, 22, 12, 34, 56),
+    )
+    assert ref.startswith("[Pasted text #1: 6 lines → ")
+    created = next(tmp_path.iterdir())
+    assert created.read_text(encoding="utf-8") == "alpha\nbeta\ngamma\ndelta\nepsilon\nzeta"
+
+
+def test_expand_paste_references_expands_exact_reference(tmp_path):
+    ref = write_pasted_text_reference(
+        "big\nchunk\nof\ntext\nfor\nagent",
+        paste_dir=tmp_path,
+        counter=2,
+        now=datetime(2026, 3, 22, 12, 34, 56),
+    )
+    assert expand_paste_references(ref) == "big\nchunk\nof\ntext\nfor\nagent"
+
+
+def test_expand_paste_references_expands_inline_reference(tmp_path):
+    ref = write_pasted_text_reference(
+        "embedded\npaste\ncontent\nline4\nline5\nline6",
+        paste_dir=tmp_path,
+        counter=3,
+        now=datetime(2026, 3, 22, 12, 34, 56),
+    )
+    text = f"Intro before\n{ref}\nOutro after"
+    expanded = expand_paste_references(text)
+    assert "embedded\npaste\ncontent" in expanded
+    assert "Intro before" in expanded
+    assert "Outro after" in expanded
+
+
+def test_expand_paste_references_leaves_missing_file_reference_literal():
+    text = "before [Pasted text #9: 6 lines → /no/such/file.txt] after"
+    assert expand_paste_references(text) == text
+
+
+def test_small_paste_into_large_existing_draft_does_not_collapse(tmp_path):
+    existing = "line1\nline2\nline3\nline4\nline5\nline6"
+    paste_dir = tmp_path / "pastes"
+    inserted, collapsed = materialize_paste_for_insertion(
+        "word",
+        current_buffer_text=existing,
+        paste_dir=paste_dir,
+        counter=1,
+        now=datetime(2026, 3, 22, 12, 34, 56),
+    )
+    assert collapsed is False
+    assert inserted == "word"
+    assert paste_dir.exists() is False
+
+
+def test_large_paste_is_inserted_as_reference_without_replacing_surrounding_text(tmp_path):
+    before = "intro\nmore intro\n"
+    after = "\noutro"
+    buf = FakeBuffer(before + after, cursor_position=len(before))
+    inserted, collapsed = materialize_paste_for_insertion(
+        "a\nb\nc\nd\ne\nf",
+        current_buffer_text=buf.text,
+        paste_dir=tmp_path,
+        counter=1,
+        now=datetime(2026, 3, 22, 12, 34, 56),
+    )
+    assert collapsed is True
+    buf.insert_text(inserted)
+    assert buf.text.startswith(before)
+    assert buf.text.endswith(after)
+    assert "[Pasted text #1:" in buf.text
+
+
+def test_large_paste_is_not_collapsed_when_current_buffer_is_slash_command(tmp_path):
+    pasted = "a\nb\nc\nd\ne\nf"
+    inserted, collapsed = materialize_paste_for_insertion(
+        pasted,
+        current_buffer_text="/plan ",
+        paste_dir=tmp_path,
+        counter=1,
+        now=datetime(2026, 3, 22, 12, 34, 56),
+    )
+    assert collapsed is False
+    assert inserted == pasted
+
+
+def test_inline_paste_reference_expands_inside_larger_message(tmp_path):
+    ref = write_pasted_text_reference(
+        "pasted\nchunk\nline3\nline4\nline5\nline6",
+        paste_dir=tmp_path,
+        counter=1,
+        now=datetime(2026, 3, 22, 12, 34, 56),
+    )
+    raw = f"Please summarize this:\n{ref}\nThanks"
+    expanded = expand_paste_references(raw)
+    assert "Please summarize this:" in expanded
+    assert "pasted\nchunk\nline3" in expanded
+    assert "Thanks" in expanded