diff --git a/amplifier_app_cli/main.py b/amplifier_app_cli/main.py
index 467b55aa..49334923 100644
--- a/amplifier_app_cli/main.py
+++ b/amplifier_app_cli/main.py
@@ -1318,7 +1318,10 @@ async def _process_runtime_mentions(session: AmplifierSession, prompt: str) -> N
         await context.add_message(msg_dict)
 
 
-def _create_prompt_session(get_active_mode: Callable | None = None) -> PromptSession:
+def _create_prompt_session(
+    get_active_mode: Callable | None = None,
+    bottom_toolbar: Callable | None = None,
+) -> PromptSession:
     """Create configured PromptSession for REPL.
 
     Provides:
@@ -1327,10 +1330,12 @@ def _create_prompt_session(get_active_mode: Callable | None = None) -> PromptSes
     - Green prompt styling matching Rich console
     - History search with Ctrl-R
     - Multi-line input with Ctrl-J
+    - Optional bottom toolbar (e.g. status bar with mode, model, session info)
     - Graceful fallback to in-memory history on errors
 
     Args:
         get_active_mode: Optional callable that returns the current active mode name
+        bottom_toolbar: Optional callable returning toolbar text (re-evaluated each render)
 
     Returns:
         Configured PromptSession instance
@@ -1391,6 +1396,7 @@ def get_prompt():
         multiline=True,  # Enable multi-line display
         prompt_continuation="  ",  # Two spaces for alignment (cleaner than "... ")
         enable_history_search=True,  # Enables Ctrl-R
+        bottom_toolbar=bottom_toolbar,
     )
 
 
@@ -1450,6 +1456,24 @@ async def interactive_chat(
 
     register_incremental_save(session, store, actual_session_id, bundle_name, config)
 
+    # Inject CLI-specific formatting policy (new sessions only — avoid duplicates on resume).
+    # Shell commands must be single-line for reliable terminal copy-paste.
+    if not session_config.is_resume:
+        cli_context = session.coordinator.get("context")
+        if cli_context and hasattr(cli_context, "add_message"):
+            await cli_context.add_message(
+                {
+                    "role": "developer",
+                    "content": (
+                        "Shell command formatting: when providing shell commands for "
+                        "the user to run, always emit each command as a single line. "
+                        "Never use backslash line continuations, heredocs, or multi-line "
+                        "constructs. If a command is long, that is acceptable — a long "
+                        "single line is easier to copy from a terminal than a multi-line one."
+                    ),
+                }
+            )
+
     # Show banner only for NEW sessions (resume shows banner via history display in commands/session.py)
     if not session_config.is_resume:
         config_summary = get_effective_config_summary(config, bundle_name)
@@ -1463,13 +1487,6 @@ async def interactive_chat(
             )
         )
 
-    # Create prompt session for history and advanced editing
-    prompt_session = _create_prompt_session(
-        get_active_mode=lambda: command_processor.session.coordinator.session_state.get(
-            "active_mode"
-        )
-    )
-
     # Helper to extract model name from config
     def _extract_model_name() -> str:
         if isinstance(config.get("providers"), list) and config["providers"]:
@@ -1481,6 +1498,47 @@ def _extract_model_name() -> str:
                 )
         return "unknown"
 
+    # Persistent footer: [mode] session | model | ~/cwd  (centered).
+    # Re-evaluated by prompt_toolkit on every keypress / redraw.
+    import html as _html
+    import shutil as _shutil
+
+    model_name = _extract_model_name()
+    short_session_id = actual_session_id[:8]
+
+    def _format_footer():
+        cols = _shutil.get_terminal_size().columns
+        mode = command_processor.session.coordinator.session_state.get("active_mode")
+
+        cwd = str(Path.cwd())
+        home = str(Path.home())
+        short_cwd = "~" + cwd[len(home) :] if cwd.startswith(home) else cwd
+
+        parts = []
+        if mode:
+            parts.append(f"<ansicyan>[{mode}]</ansicyan>")
+        parts.append(_html.escape(short_session_id))
+        parts.append(_html.escape(model_name))
+        parts.append(_html.escape(short_cwd))
+        content = " | ".join(parts)
+
+        # Center within the terminal width
+        content_len = (
+            len(short_session_id) + len(model_name) + len(short_cwd) + 3 * len(" | ")
+        )
+        if mode:
+            content_len += len(f"[{mode}]") + len(" | ")
+        pad = max(0, (cols - content_len) // 2)
+        return HTML(" " * pad + content)
+
+    # Create prompt session for history and advanced editing
+    prompt_session = _create_prompt_session(
+        get_active_mode=lambda: command_processor.session.coordinator.session_state.get(
+            "active_mode"
+        ),
+        bottom_toolbar=_format_footer,
+    )
+
     # Helper to save session after each turn
     async def _save_session():
         context = session.coordinator.get("context")
diff --git a/amplifier_app_cli/ui/message_renderer.py b/amplifier_app_cli/ui/message_renderer.py
index 1b9734d2..ecfae64a 100644
--- a/amplifier_app_cli/ui/message_renderer.py
+++ b/amplifier_app_cli/ui/message_renderer.py
@@ -6,12 +6,101 @@
 Zero duplication: All message rendering goes through these functions.
 """
 
+import re
+
+from markdown_it import MarkdownIt
 from rich.console import Console
 
 from ..console import Markdown
 
+_SHELL_LEXERS = frozenset({"bash", "sh", "shell", "zsh", "fish", "console"})
+
+# Shared markdown-it parser for extracting code fence tokens.
+# This correctly handles backticks inside fenced blocks (unlike regex).
+_md_parser = MarkdownIt()
+
+# Fixed-width rule (72 chars) for command block framing.
+_RULE_WIDTH = 72
+_RULE = "\u2500" * _RULE_WIDTH  # ────────...
+_SCISSORS = "\u2702"  # ✂
+
+
+def _print_framed_command(code: str, console: Console) -> None:
+    """Print a shell command with visual framing for copy-paste.
+
+    Renders:
+        ───── ✂ Copy & Run ──────────────────────────────────────
+        <blank line>
+        <command as single logical line, soft_wrap=True>
+        <blank line>
+        ─────────────────────────────────────────────────────────
+
+    The command is plain text with soft_wrap so the terminal handles visual
+    wrapping. Triple-click copies the full command as one logical line.
+    """
+    # Top rule with scissors label
+    label = f" {_SCISSORS} Copy & Run "
+    top_rule = "\u2500" * 5 + label + "\u2500" * max(1, _RULE_WIDTH - 5 - len(label))
+    console.print(f"\n[dim]{top_rule}[/dim]")
+
+    # Command — plain text, single logical line, no Rich highlighting
+    console.print()
+    console.print(code, soft_wrap=True, highlight=False)
+    console.print()
+
+    # Bottom rule
+    console.print(f"[dim]{_RULE}[/dim]")
+
+
+def _render_content_with_copyable_commands(content: str, console: Console) -> None:
+    """Render markdown, framing shell code blocks for copy-paste.
+
+    Uses markdown-it-py to parse the AST — this correctly identifies code fence
+    boundaries even when the command content contains backticks (e.g. commands
+    that generate markdown with echo '```'). Regex fence matching breaks on
+    these cases.
+
+    Shell code blocks get visual framing (scissors rule + plain text + bottom
+    rule). Non-shell code blocks render normally through Rich Markdown.
+    """
+    tokens = _md_parser.parse(content)
+    lines = content.split("\n")
+
+    # Collect shell fence tokens with their line ranges
+    shell_fences: list[tuple[int, int, str]] = []
+    for token in tokens:
+        if token.type == "fence" and token.map:
+            lang = ((token.info or "").split() or [""])[0]
+            if lang in _SHELL_LEXERS:
+                shell_fences.append((token.map[0], token.map[1], token.content))
+
+    if not shell_fences:
+        # No shell blocks — render everything through Markdown
+        console.print(Markdown(content))
+        return
+
+    last_line = 0
+    for start_line, end_line, code in shell_fences:
+        # Render any markdown content before this shell block
+        before = "\n".join(lines[last_line:start_line])
+        if before.strip():
+            console.print(Markdown(before))
+
+        # Collapse backslash continuations and frame the command
+        code = re.sub(r" *\\\n[ \t]*", " ", code).rstrip()
+        _print_framed_command(code, console)
+
+        last_line = end_line
+
+    # Render any remaining content after the last shell block
+    remaining = "\n".join(lines[last_line:])
+    if remaining.strip():
+        console.print(Markdown(remaining))
+
 
-def render_message(message: dict, console: Console, *, show_thinking: bool = False) -> None:
+def render_message(
+    message: dict, console: Console, *, show_thinking: bool = False
+) -> None:
     """Render a single message (user or assistant).
 
     Single source of truth for message formatting. Used by:
@@ -39,9 +128,13 @@ def _render_user_message(message: dict, console: Console) -> None:
     console.print(f"\n[bold green]>[/bold green] {content}")
 
 
-def _render_assistant_message(message: dict, console: Console, show_thinking: bool) -> None:
+def _render_assistant_message(
+    message: dict, console: Console, show_thinking: bool
+) -> None:
     """Render assistant message with green prefix and markdown."""
-    text_blocks, thinking_blocks = _extract_content_blocks(message, show_thinking=show_thinking)
+    text_blocks, thinking_blocks = _extract_content_blocks(
+        message, show_thinking=show_thinking
+    )
 
     # Skip rendering if message is empty (tool-only messages)
     if not text_blocks and not thinking_blocks:
@@ -49,16 +142,20 @@ def _render_assistant_message(message: dict, console: Console, show_thinking: bo
 
     console.print("\n[bold green]Amplifier:[/bold green]")
 
-    # Render text blocks with default styling
+    # Render text blocks — shell code blocks get framing for copy-paste.
+    # _render_content_with_copyable_commands parses the AST once and
+    # short-circuits to plain Markdown when no shell fences are found.
     if text_blocks:
-        console.print(Markdown("\n".join(text_blocks)))
+        _render_content_with_copyable_commands("\n".join(text_blocks), console)
 
     # Render thinking blocks with dim styling
     for thinking in thinking_blocks:
-        console.print(Markdown(f"\n💭 **Thinking:**\n{thinking}", style="dim"))
+        console.print(Markdown(f"\n\U0001f4ad **Thinking:**\n{thinking}", style="dim"))
 
 
-def _extract_content_blocks(message: dict, *, show_thinking: bool = False) -> tuple[list[str], list[str]]:
+def _extract_content_blocks(
+    message: dict, *, show_thinking: bool = False
+) -> tuple[list[str], list[str]]:
     """Extract text and thinking blocks separately from message content.
 
     Handles multiple content formats:
@@ -123,7 +220,7 @@ def _extract_content(message: dict, *, show_thinking: bool = False) -> str:
                 text_parts.append(block.get("text", ""))
             elif block.get("type") == "thinking" and show_thinking:
                 thinking = block.get("thinking", "")
-                text_parts.append(f"\n[dim]💭 Thinking: {thinking}[/dim]\n")
+                text_parts.append(f"\n[dim]\U0001f4ad Thinking: {thinking}[/dim]\n")
         return "\n".join(text_parts)
 
     # Fallback for unexpected formats
diff --git a/uv.lock b/uv.lock
index 6b149161..011dab57 100644
--- a/uv.lock
+++ b/uv.lock
@@ -13,6 +13,7 @@ dependencies = [
     { name = "httpx", extra = ["socks"] },
     { name = "prompt-toolkit" },
     { name = "pydantic" },
+    { name = "pygments" },
     { name = "pyyaml" },
     { name = "rich" },
 ]
@@ -25,12 +26,13 @@ dev = [
 
 [package.metadata]
 requires-dist = [
-    { name = "amplifier-core", git = "https://github.com/microsoft/amplifier-core?branch=main" },
+    { name = "amplifier-core" },
     { name = "amplifier-foundation", git = "https://github.com/microsoft/amplifier-foundation?branch=main" },
     { name = "click", specifier = ">=8.1.0" },
     { name = "httpx", extras = ["socks"], specifier = ">=0.28.1" },
     { name = "prompt-toolkit", specifier = ">=3.0.52" },
     { name = "pydantic", specifier = ">=2.0.0" },
+    { name = "pygments", specifier = ">=2.13.0" },
     { name = "pyyaml", specifier = ">=6.0.3" },
     { name = "rich", specifier = ">=13.0.0" },
 ]