fix: voice loop — persistent event loop, markdown stripping, MCP noise

Three fixes from real-world testing:

1. Event loop: replaced asyncio.run() with a persistent loop so
   Agno's MCP sessions survive across conversation turns. No more
   'Event loop is closed' errors on turn 2+.

2. Markdown stripping: voice preamble tells Timmy to respond in
   natural spoken language, plus _strip_markdown() as a safety net
   removes **bold**, *italic*, bullets, headers, code fences, etc.
   TTS no longer reads 'asterisk asterisk'.

3. MCP noise: _suppress_mcp_noise() quiets mcp/agno/httpx loggers
   during voice mode so the terminal shows clean transcript only.

32 tests (12 new for markdown stripping + persistent loop).

src/timmy/voice_loop.py

@@ -15,6 +15,7 @@ Requires: sounddevice, numpy, whisper, piper-tts
 
 import asyncio
 import logging
+import re
 import subprocess
 import sys
 import tempfile
@@ -26,6 +27,44 @@ import numpy as np
 
 logger = logging.getLogger(__name__)
 
+# ── Voice-mode system instruction ───────────────────────────────────────────
+# Prepended to user messages so Timmy responds naturally for TTS.
+_VOICE_PREAMBLE = (
+    "[VOICE MODE] You are speaking aloud through a text-to-speech system. "
+    "Respond in short, natural spoken sentences. No markdown, no bullet points, "
+    "no asterisks, no numbered lists, no headers, no bold/italic formatting. "
+    "Talk like a person in a conversation — concise, warm, direct. "
+    "Keep responses under 3-4 sentences unless the user asks for detail."
+)
+
+
+def _strip_markdown(text: str) -> str:
+    """Remove markdown formatting so TTS reads naturally.
+
+    Strips: **bold**, *italic*, `code`, # headers, - bullets,
+    numbered lists, [links](url), etc.
+    """
+    if not text:
+        return text
+    # Remove bold/italic markers
+    text = re.sub(r"\*{1,3}([^*]+)\*{1,3}", r"\1", text)
+    # Remove inline code
+    text = re.sub(r"`([^`]+)`", r"\1", text)
+    # Remove headers (# Header)
+    text = re.sub(r"^#{1,6}\s+", "", text, flags=re.MULTILINE)
+    # Remove bullet points (-, *, +) at start of line
+    text = re.sub(r"^[\s]*[-*+]\s+", "", text, flags=re.MULTILINE)
+    # Remove numbered lists (1. 2. etc)
+    text = re.sub(r"^[\s]*\d+\.\s+", "", text, flags=re.MULTILINE)
+    # Remove link syntax [text](url) → text
+    text = re.sub(r"\[([^\]]+)\]\([^)]+\)", r"\1", text)
+    # Remove horizontal rules
+    text = re.sub(r"^[-*_]{3,}\s*$", "", text, flags=re.MULTILINE)
+    # Collapse multiple newlines
+    text = re.sub(r"\n{3,}", "\n\n", text)
+    return text.strip()
+
+
 # ── Defaults ────────────────────────────────────────────────────────────────
 
 DEFAULT_WHISPER_MODEL = "base.en"
@@ -75,6 +114,9 @@ class VoiceLoop:
         self._running = False
         self._speaking = False  # True while TTS is playing
         self._interrupted = False  # set when user talks over TTS
+        # Persistent event loop — reused across all chat calls so Agno's
+        # MCP sessions don't die when the loop closes.
+        self._loop: asyncio.AbstractEventLoop | None = None
 
     # ── Lazy initialization ─────────────────────────────────────────────
 
@@ -283,6 +325,16 @@ class VoiceLoop:
 
     # ── LLM: Text → Response ───────────────────────────────────────────
 
+    def _get_loop(self) -> asyncio.AbstractEventLoop:
+        """Return a persistent event loop, creating one if needed.
+
+        A single loop is reused for the entire voice session so Agno's
+        MCP tool-server connections survive across turns.
+        """
+        if self._loop is None or self._loop.is_closed():
+            self._loop = asyncio.new_event_loop()
+        return self._loop
+
     def _think(self, user_text: str) -> str:
         """Send text to Timmy and get a response."""
         sys.stdout.write("  💭 Thinking...\r")
@@ -291,20 +343,29 @@ class VoiceLoop:
         t0 = time.monotonic()
 
         try:
-            response = asyncio.run(self._chat(user_text))
+            loop = self._get_loop()
+            response = loop.run_until_complete(self._chat(user_text))
         except Exception as exc:
             logger.error("Timmy chat failed: %s", exc)
             response = "I'm having trouble thinking right now. Could you try again?"
 
         elapsed = time.monotonic() - t0
         logger.info("Timmy responded in %.1fs", elapsed)
+
+        # Strip markdown so TTS doesn't read asterisks, bullets, etc.
+        response = _strip_markdown(response)
         return response
 
     async def _chat(self, message: str) -> str:
-        """Async wrapper around Timmy's session.chat()."""
+        """Async wrapper around Timmy's session.chat().
+
+        Prepends the voice-mode instruction so Timmy responds in
+        natural spoken language rather than markdown.
+        """
         from timmy.session import chat
 
-        return await chat(message, session_id=self.config.session_id)
+        voiced = f"{_VOICE_PREAMBLE}\n\nUser said: {message}"
+        return await chat(voiced, session_id=self.config.session_id)
 
     # ── Main Loop ───────────────────────────────────────────────────────
 
@@ -312,6 +373,11 @@ class VoiceLoop:
         """Run the voice loop. Blocks until Ctrl-C."""
         self._ensure_piper()
 
+        # Suppress MCP / Agno stderr noise during voice mode.
+        # The "Secure MCP Filesystem Server running on stdio" messages
+        # are distracting in a voice session.
+        _suppress_mcp_noise()
+
         tts_label = (
             "macOS say"
             if self.config.use_say_fallback
@@ -381,7 +447,36 @@ class VoiceLoop:
             print("\n\n  👋 Voice loop stopped.\n")
         finally:
             self._running = False
+            self._cleanup_loop()
+
+    def _cleanup_loop(self) -> None:
+        """Shut down the persistent event loop cleanly."""
+        if self._loop is not None and not self._loop.is_closed():
+            try:
+                self._loop.run_until_complete(self._loop.shutdown_asyncgens())
+            except Exception:
+                pass
+            self._loop.close()
+            self._loop = None
 
     def stop(self) -> None:
         """Stop the voice loop (from another thread)."""
         self._running = False
+
+
+def _suppress_mcp_noise() -> None:
+    """Quiet down noisy MCP/Agno loggers during voice mode.
+
+    Sets specific loggers to WARNING so the terminal stays clean
+    for the voice transcript.
+    """
+    for name in (
+        "mcp",
+        "mcp.server",
+        "mcp.client",
+        "agno",
+        "agno.mcp",
+        "httpx",
+        "httpcore",
+    ):
+        logging.getLogger(name).setLevel(logging.WARNING)