diff --git a/clients/terminal/cli.py b/clients/terminal/cli.py
index 25b613d..bdb10bc 100644
--- a/clients/terminal/cli.py
+++ b/clients/terminal/cli.py
@@ -283,16 +283,18 @@
         try:
             session = await api.get_session(target)
         except Exception as exc:
-            # Try to find by prefix
+            # Exact id miss — try a prefix match. If there is not exactly one,
+            # surface the original error too (404 vs network vs server) rather
+            # than swallowing it into a generic "not found".
             try:
                 sessions = await api.list_sessions()
                 matches = [s for s in sessions if s["session_id"].startswith(target)]
-                if len(matches) == 1:
-                    session = matches[0]
-                else:
-                    raise exc
             except Exception:
-                click.secho(f"Session not found: {target}", fg="red", err=True)
+                matches = []
+            if len(matches) == 1:
+                session = matches[0]
+            else:
+                click.secho(f"Session not found: {target} ({exc})", fg="red", err=True)
                 return True
         state.set_session_id(session["session_id"])
         click.secho(
diff --git a/clients/terminal/tui/file_refs.py b/clients/terminal/tui/file_refs.py
index 34ca6f5..7e58886 100644
--- a/clients/terminal/tui/file_refs.py
+++ b/clients/terminal/tui/file_refs.py
@@ -274,9 +274,12 @@
 def _collect_files(path: Path, recursive: bool = False) -> Iterable[Path]:
     """Yield non-sensitive files inside a directory."""
     iterator = path.rglob("*") if recursive else path.iterdir()
-    for p in sorted(iterator):
-        if p.is_file() and not _is_inside_sensitive_dir(p):
-            yield p
+    # Filter sensitive entries in the generator BEFORE sorted(), so a huge
+    # directory (deep project tree) does not get fully materialized into a list
+    # only to have most of it discarded — sort only the kept files.
+    kept = (p for p in iterator if p.is_file() and not _is_inside_sensitive_dir(p))
+    for p in sorted(kept):
+        yield p
 
 
 def _display_path(path: Path, base: Path, root_dir: Path | None = None) -> str:
diff --git a/clients/terminal/tui/renderers/filesystem.py b/clients/terminal/tui/renderers/filesystem.py
index 28f21e3..d51247c 100644
--- a/clients/terminal/tui/renderers/filesystem.py
+++ b/clients/terminal/tui/renderers/filesystem.py
@@ -360,11 +360,17 @@
             return Text(text, style=theme.text_dim.hex)
 
         pattern = str(args.get("pattern", "") or "")
-        # Pre-compile a case-insensitive literal match for highlighting.
+        # Pre-compile a match for highlighting. Literal by default; when the
+        # grep ran in regex mode, highlight with the actual regex so the marked
+        # spans match what the server found (a literal highlight of a regex
+        # pattern would mark nothing, diverging from the real matches).
         highlight = None
         if pattern:
             try:
-                highlight = re.compile(re.escape(pattern), re.IGNORECASE)
+                if args.get("regex"):
+                    highlight = re.compile(pattern, re.IGNORECASE)
+                else:
+                    highlight = re.compile(re.escape(pattern), re.IGNORECASE)
             except re.error:
                 highlight = None
 
diff --git a/clients/terminal/tui/renderers/tool.py b/clients/terminal/tui/renderers/tool.py
index 16bffe2..f513489 100644
--- a/clients/terminal/tui/renderers/tool.py
+++ b/clients/terminal/tui/renderers/tool.py
@@ -15,6 +15,20 @@
 
 from .base import ContentRenderer
 
+# Cap a generic tool result so a tool with huge output (non-filesystem — those
+# have their own renderer) does not flood the chat bubble. Keep the tail, with
+# a marker, mirroring shell_runner's truncation.
+_RESULT_MAX_LINES = 200
+
+
+def _truncate_result(text: str) -> str:
+    """Keep the last ``_RESULT_MAX_LINES`` lines of a tool result, with a marker."""
+    lines = text.splitlines()
+    if len(lines) <= _RESULT_MAX_LINES:
+        return text
+    dropped = len(lines) - _RESULT_MAX_LINES
+    return f"... [{dropped} lines truncated]\n" + "\n".join(lines[-_RESULT_MAX_LINES:])
+
 
 class ToolStartedRenderer(ContentRenderer):
     """Render a tool call start marker.
@@ -67,7 +81,10 @@
         result = msg.get("result")
         color = theme.tool_success if success else theme.tool_error
         title = f"← {tool} {'✓' if success else '✗'}"
-        body = Text(str(result) if result is not None else "", style=theme.text_dim.hex)
+        body = Text(
+            _truncate_result(str(result) if result is not None else ""),
+            style=theme.text_dim.hex,
+        )
         panel = Panel(
             body,
             title=title,
diff --git a/clients/terminal/tui/tui_app.py b/clients/terminal/tui/tui_app.py
index 24df484..36e4166 100644
--- a/clients/terminal/tui/tui_app.py
+++ b/clients/terminal/tui/tui_app.py
@@ -174,7 +174,7 @@
             # is the lenient path.
             try:
                 session = await api.get_session(session_id)
-                return session["session_id"]
+                return session.get("session_id")
             except Exception as exc:
                 self._chat_panel.handle_ws_event(
                     {"type": "error", "message": f"Failed to resume session {session_id[:8]}: {exc}"}
@@ -186,7 +186,7 @@
             if saved:
                 try:
                     session = await api.get_session(saved)
-                    return session["session_id"]
+                    return session.get("session_id")
                 except Exception:
                     self._state.clear_session_id()
 
@@ -198,8 +198,10 @@
                 {"type": "error", "message": f"Failed to create session: {exc}"}
             )
             return None
-        self._state.set_session_id(session["session_id"])
-        return session["session_id"]
+        sid = session.get("session_id")
+        if sid:
+            self._state.set_session_id(sid)
+        return sid
 
     async def attach_session(self, session_id: str) -> None:
         self._ctx.session_id = session_id
diff --git a/clients/terminal/tui/widgets/input_box.py b/clients/terminal/tui/widgets/input_box.py
index 8eb42af..7cfcee6 100644
--- a/clients/terminal/tui/widgets/input_box.py
+++ b/clients/terminal/tui/widgets/input_box.py
@@ -85,6 +85,12 @@
         if hints is not None and hints.visible():
             cmd = hints.current_match()
         if cmd is None:
+            # A bare "/" with no command name typed: do not auto-pick the first
+            # command in the registry — the user has chosen nothing. (If the
+            # hints list is open, current_match above already reflects their
+            # highlighted choice.)
+            if not text[1:].strip():
+                return False
             from clients.terminal.tui.commands.registry import get_registry
 
             matches = get_registry().match(text[1:])
diff --git a/tests/clients/test_file_refs.py b/tests/clients/test_file_refs.py
index 4611e67..28de1a1 100644
--- a/tests/clients/test_file_refs.py
+++ b/tests/clients/test_file_refs.py
@@ -48,6 +48,24 @@
     assert "sub/util.py" in paths
 
 
+def test_resolve_directory_skips_sensitive_subdir(sample_dir: Path) -> None:
+    """A sensitive subdir (e.g. .venv) is filtered out before sort, so its
+    files never get attached (regression for 5.P1)."""
+    (sample_dir / "sub" / "deep.py").write_text("deep\n", encoding="utf-8")
+    venv = sample_dir / ".venv"
+    venv.mkdir()
+    (venv / "site.py").write_text("should be skipped\n", encoding="utf-8")
+
+    resolver = FileRefResolver(sample_dir)
+    result = resolver.resolve("look @./")
+    paths = {a.display_path for a in result.attachments}
+    assert "main.py" in paths
+    assert "sub/util.py" in paths
+    assert "sub/deep.py" in paths
+    assert ".venv/site.py" not in paths
+    assert not any(".venv" in p for p in paths)
+
+
 def test_resolve_missing_file(sample_dir: Path) -> None:
     resolver = FileRefResolver(sample_dir)
     result = resolver.resolve("check @missing.py")
diff --git a/tests/clients/test_filesystem_renderer.py b/tests/clients/test_filesystem_renderer.py
index 798bce2..967c071 100644
--- a/tests/clients/test_filesystem_renderer.py
+++ b/tests/clients/test_filesystem_renderer.py
@@ -360,6 +360,34 @@
     assert theme.accent.hex in styles_foo_bar
 
 
+def test_grep_regex_mode_highlights_regex_matches() -> None:
+    """When the grep ran in regex mode, the highlight uses the actual regex so
+    the marked spans match what the server found — a literal highlight of a
+    regex pattern would mark nothing (regression for 2.B2)."""
+    set_active_theme("gnexus-dark")
+    theme = get_active_theme()
+    renderer = FilesystemToolResultRenderer()
+    msg = {
+        "type": "tool_call",
+        "tool": "filesystem",
+        "args": {"action": "grep", "pattern": r"\d+", "regex": True},
+        "result": (
+            "[2 matches for '\\d+' in /proj (1 files searched)]\n"
+            "src/a.py:3: value 42 here\n"
+            "src/b.py:7: count 7 total\n",
+        ),
+        "success": True,
+    }
+    body = _body(renderer.render(msg))
+    assert isinstance(body, Text)
+    # The numbers (regex matches) are highlighted in accent; the literal "\d+"
+    # text in the header plaque is dim (not matched as a literal).
+    styles_42 = _span_styles_at(body, "42")
+    assert theme.accent.hex in styles_42
+    styles_count7 = _span_styles_at(body, "7 total")
+    assert theme.accent.hex in styles_count7
+
+
 def test_grep_no_matches_renders_dim() -> None:
     set_active_theme("gnexus-dark")
     theme = get_active_theme()
diff --git a/tests/clients/test_input_box.py b/tests/clients/test_input_box.py
index 28ae9eb..73479db 100644
--- a/tests/clients/test_input_box.py
+++ b/tests/clients/test_input_box.py
@@ -112,6 +112,16 @@
         assert input_box._input.text == "/switch "
 
 
+def test_tab_bare_slash_without_hints_does_not_auto_pick() -> None:
+    """A bare ``/`` with no hint list available must not auto-complete to the
+    first command in the registry — the user has chosen nothing (3.B3)."""
+    from clients.terminal.tui.widgets.input_box import _PromptInput
+
+    p = _PromptInput(text="/", hints=None)
+    assert p._complete_command() is False
+    assert p.text == "/"
+
+
 @pytest.mark.anyio
 async def test_tab_does_not_complete_without_slash() -> None:
     async with NaviCodeTui(new_session=True).run_test() as pilot:
diff --git a/tests/clients/test_terminal_client.py b/tests/clients/test_terminal_client.py
index 9fc6951..95b3c11 100644
--- a/tests/clients/test_terminal_client.py
+++ b/tests/clients/test_terminal_client.py
@@ -237,3 +237,30 @@
         result = runner.invoke(main, ["--resume", "sess-x", "--new-session"])
         assert result.exit_code != 0
         assert "mutually exclusive" in result.output
+
+    def test_switch_surfaces_original_error_on_no_match(
+        self, monkeypatch: pytest.MonkeyPatch, tmp_path: Path, capsys: pytest.CaptureFixture
+    ) -> None:
+        """/switch with no exact + no unique prefix shows the original error cause,
+        not just a generic "not found" (regression for 6.B2)."""
+        import asyncio
+
+        import clients.terminal.api as api_module
+
+        async def boom(_sid: str) -> dict:
+            raise RuntimeError("server 500")
+
+        async def no_sessions() -> list[dict]:
+            return []
+
+        monkeypatch.setattr(api_module, "get_session", boom)
+        monkeypatch.setattr(api_module, "list_sessions", no_sessions)
+
+        from clients.terminal.cli import _handle_command
+
+        state = StateManager(tmp_path)
+        state.set_session_id("sess-current")
+        asyncio.run(_handle_command("/switch zzz", state, client=object()))
+        out = capsys.readouterr().err
+        assert "Session not found: zzz" in out
+        assert "server 500" in out  # original error preserved
diff --git a/tests/clients/test_tool_result_renderer.py b/tests/clients/test_tool_result_renderer.py
new file mode 100644
index 0000000..3f017f1
--- /dev/null
+++ b/tests/clients/test_tool_result_renderer.py
@@ -0,0 +1,50 @@
+"""Tests for the generic tool result renderer."""
+
+from __future__ import annotations
+
+from rich.console import Console
+
+from clients.terminal.tui.renderers.tool import ToolResultRenderer
+from clients.terminal.tui.themes import set_active_theme
+
+
+def _render_text(renderable) -> str:
+    console = Console(record=True, width=80, force_terminal=True, color_system=None)
+    console.print(renderable)
+    return console.export_text()
+
+
+def test_tool_result_truncates_long_output() -> None:
+    """A non-filesystem tool result with more than the line cap is truncated to
+    the tail with a marker, so a huge output does not flood the chat bubble
+    (regression for 2.P3)."""
+    set_active_theme("gnexus-dark")
+    renderer = ToolResultRenderer()
+    big = "\n".join(f"line {i}" for i in range(300))
+    msg = {
+        "type": "tool_call",
+        "tool": "some_tool",
+        "result": big,
+        "success": True,
+    }
+    panel = renderer.render(msg)
+    text = _render_text(panel.renderable)
+    assert "lines truncated]" in text
+    # The tail is kept (last line present), the head is dropped (first line not).
+    assert "line 299" in text
+    assert "line 0" not in text
+
+
+def test_tool_result_short_output_not_truncated() -> None:
+    set_active_theme("gnexus-dark")
+    renderer = ToolResultRenderer()
+    msg = {
+        "type": "tool_call",
+        "tool": "some_tool",
+        "result": "just a few\nlines",
+        "success": True,
+    }
+    panel = renderer.render(msg)
+    text = _render_text(panel.renderable)
+    assert "truncated" not in text
+    assert "just a few" in text
\ No newline at end of file
diff --git a/tests/clients/test_tui_app.py b/tests/clients/test_tui_app.py
index 148c98e..eef554f 100644
--- a/tests/clients/test_tui_app.py
+++ b/tests/clients/test_tui_app.py
@@ -572,6 +572,24 @@
 
 
 @pytest.mark.anyio
+async def test_resolve_session_returns_none_when_response_lacks_session_id(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    """A malformed server response (no session_id) must not KeyError — returns
+    None so the caller skips attaching (regression for 5.B3)."""
+    import clients.terminal.api as api_module
+
+    async def fake_get_session(sid):
+        return {"profile_id": "navi_code"}  # no session_id key
+
+    monkeypatch.setattr(api_module, "get_session", fake_get_session)
+    async with NaviCodeTui(new_session=True).run_test() as pilot:
+        await pilot.pause()
+        result = await pilot.app._resolve_session("sess-x", None, False)
+        assert result is None
+
+
+@pytest.mark.anyio
 async def test_activity_indicator_runs_during_turn() -> None:
     """stream_start starts the spinner; stream_end stops it."""
     async with NaviCodeTui(new_session=True).run_test() as pilot: