feat: start skips the mic check by default; --check to opt in

invert the pre-listen mic check — default is no check (just start listening); pass
'claudedo start --check' to run it. replaces the old --skip-audio-check flag.

Signed-off-by: disqualifier <dev@disqualifier.me>

.gitignore

@@ -1,4 +1,5 @@
 CLAUDE.md
+COMPACT.md
 
 __pycache__/
 *.pyc

README.md

@@ -11,7 +11,7 @@ hands-free while another window (a game) is focused.
 It exists because Claude Code's native `/voice` is hardcoded-blocked in WSL (it
 assumes WSL has no audio). Modern WSL2 + WSLg *does* have working mic input via
 PulseAudio/RDP. `claudedo` captures the mic itself, transcribes on-device, and drives
-Claude Code over tmux — fully local, private, backgroundable.
+Claude Code over tmux — fully local and private. You run it in a terminal you watch.
 
 ## How it works
 
@@ -20,9 +20,11 @@ mic (WSLg/PulseAudio RDPSource)
   -> sounddevice capture
   -> faster-whisper (local STT, on-device)
   -> wake gate: utterance must start with a wake phrase, else DISCARD locally
-  -> grammar match (yes/no/one..four/approve/deny/send/type/mode/set/target/cancel)
+  -> grammar match (yes/no/one..four/approve/deny/send/type/space/backspace/erase/
-  -> resolve target session (~/.claude-active)
+                    mode/set/target/unset/list/cancel)
+  -> resolve target session (one-shot > sticky ~/.claude-active > auto/none)
   -> tmux send-keys -t <session> "<keys>"
+  -> log the action to the watched terminal ([session]/[SYSTEM]/[VOICE], colored)
 ```
 
 **Privacy by construction.** STT runs on-device. In listen mode, any speech that
@@ -62,16 +64,19 @@ claudedo test-audio
 ## Usage
 
 **Run it in a terminal you watch — that's the product.** You launch `claudedo
-start`, it does a quick mic check, then drops into a visible listen loop that prints
+start` and it drops into a visible listen loop (pass `--check` to run a mic check
-`heard → matched → sent` for every utterance. That terminal is your
+first). Each utterance prints a timestamped, colored line — `HH:MM:SS [claude-libs]
-recognition/action console; you attach to the `claude-<name>` session in another pane
+heard "…" →
-to watch the keystrokes land. There is no backgrounding/daemon mode — the whole point
+typed 'fix'` (green for injected, red for drops, `[SYSTEM]`/`[VOICE]` for state and
-is the console you read.
+recognition). That terminal is your recognition/action console; you attach to the
+`claude-<name>` session in another pane to watch the keystrokes land. It runs in the
+foreground by design — the console is the point — though `claudedo stop` can signal a
+stray instance.
 
 ```bash
-claudedo start            # mic-check, then the visible listen loop (listen mode default)
+claudedo start            # the visible listen loop (listen mode default; no mic check)
+claudedo start --check    # run a mic check before listening
 claudedo start --mode ptt # push-to-talk instead (desk-only — see Modes)
-claudedo start --skip-audio-check  # skip the pre-listen mic check
 claudedo status           # running? mode? target session?
 claudedo stop             # stop a running daemon
 claudedo set <name>       # set the sticky target -> claude-<name> (alias: switch)
@@ -97,9 +102,12 @@ Switch at runtime by voice: "claudedo mode listen" / "claudedo mode ptt".
 
 ## Command grammar
 
-Wake phrases (listen mode), fuzzy-matched: **"claudedo"**, **"hey claude"**.
+Wake phrases (listen mode), fuzzy-matched. The default list is **"claudedo"**,
-"claudedo" is a coined word, so the matcher is lenient (accepts "claude do",
+**"claude do"**, **"hey claude"**, **"ok claude"**, **"okay claude"** — Whisper has
-"clauddo", "cloud do", …). In PTT mode the wake phrase is optional.
+no token for the coined word "claudedo" and renders it as real words ("claude do"),
+so that spelling is listed explicitly. Matching is lenient (case/space-insensitive).
+Add the spellings you actually see (turn on `print_heard` to find them). In PTT mode
+the wake phrase is optional.
 
 | Say | Does |
 |---|---|
@@ -108,6 +116,10 @@ Wake phrases (listen mode), fuzzy-matched: **"claudedo"**, **"hey claude"**.
 | `approve` / `deny` | allow / deny a permission prompt |
 | `send` / `enter` | submit (Enter) |
 | `type <phrase>` | insert literal text, **no** submit (read-before-send; say "send") |
+| `space [<n>]` (also `add [a] space`, `insert <n> spaces`) | insert n spaces (default 1) |
+| `backspace [<n>]` (alias `delete`) | delete n chars (default 1), capped at the last submit boundary |
+| `erase` (alias `clear`/`wipe`) | delete everything typed since the last submit/boundary |
+| `debug <text>` (alias `echo`) | just print what you said to the console (test wake/STT; injects nothing) |
 | `mode ptt` / `mode listen` | switch input mode |
 | `set <name>` (alias `sticky`/`switch`) | set the **sticky** target → `claude-<name>` (persists) |
 | `target <name> <command>` | **one-shot** override: run that command on `claude-<name>` for this utterance only; sticky default unchanged |
@@ -118,8 +130,10 @@ Wake phrases (listen mode), fuzzy-matched: **"claudedo"**, **"hey claude"**.
 Optional filler (`select` / `use` / `choose`) may precede any command and is ignored:
 `select yes` and `use yes` behave like `yes`. (`select 1` is still the select command.)
 
-When no sticky target is set, a bare command auto-targets the **only** running
+When no sticky target is set, a bare command does nothing and asks you to `set` one
-`claude-*` session; if several are running it does nothing and asks you to `set` one.
+(the default). Set `auto_target = true` to instead auto-use the single running
+`claude-*` session when there's exactly one; with several running it always does
+nothing and asks you to `set` one.
 
 Number words are normalized to digits before matching ("one"/"won" → 1).
 
@@ -132,9 +146,9 @@ A `target <name>` voice command is a **one-shot** that does NOT touch the sticky
 default — it routes a single command and the next bare command reverts to sticky.
 
 Resolution order (one place — `target.resolve()`): one-shot if present →
-sticky if set and the session exists → else the only running `claude-*` session →
+sticky if set and the session exists → else, only if `auto_target = true`, the single
-else (zero or several) do nothing and say so. It never guesses, and never injects
+running `claude-*` session → else (default, or zero/several sessions) do nothing and
-into a nonexistent session.
+say so. It never guesses, and never injects into a nonexistent session.
 
 Every name maps to `claude-<name>` through one helper (`target.session_name()`), and
 the cc kit mirrors it exactly — so `cc libs` (shell) and `set libs` (voice) refer
@@ -171,11 +185,29 @@ If Claude Code changes its prompt UI, re-confirm against a live session and upda
 ## Config
 
 Everything tunable lives in [`config.toml`](config.toml): wake phrases, mode + PTT
-key, Whisper model/language/device, audio segmentation thresholds, and
+key, Whisper model/language/device, `[vad]` endpointing, and `[behavior]`
-`type_autosend = false`. The default model is `small`; bump to `medium` if the coined
+(`type_autosend`, fuzzy thresholds, `filler_words`, `auto_target`, `print_heard`).
-wake word is recognized poorly. `claudedo -c <path> ...` points at a specific config;
+The default model is **`medium`** (best accuracy for the coined wake word on a strong
-otherwise it searches `$CLAUDEDO_CONFIG`, `~/.config/claudedo/config.toml`, then
+CPU); `small` is faster/less accurate, `large-v3` most accurate. `claudedo -c <path>
-`./config.toml`.
+...` points at a specific config; otherwise it searches `$CLAUDEDO_CONFIG`,
+`~/.config/claudedo/config.toml`, then `./config.toml`.
+
+- **STT biasing.** The transcriber is seeded with an `initial_prompt` built from the
+  configured wake phrases + command vocabulary (one source — `grammar.vocabulary()`),
+  so Whisper is conditioned to expect "claudedo" and the command words.
+- **Split fuzzy thresholds.** `wake_fuzzy_threshold` (default `0.6`, lenient) vs
+  `command_fuzzy_threshold` (default `0.8`, tight). The asymmetry is deliberate: a
+  false *wake* is cheap (it wakes, finds no command, does nothing), but a false
+  *command* fires the wrong action. Prefer expanding command synonyms over loosening
+  the command threshold.
+- **`[vad]` endpointing.** Capture starts on speech and ends after `silence_ms`
+  (default 800) of trailing silence — Alexa-style record-until-pause — capped at
+  `max_seconds` (default 10). The pause both ends a command and separates it from
+  following chatter (the chatter is a separate capture the wake gate discards).
+- **`auto_target`** (default `false`): with no sticky target and one session running,
+  `false` does nothing and asks you to `set`; `true` auto-uses that session.
+- **`print_heard`** (default `false`, debug): prints non-wake transcripts so you can
+  see how Whisper renders your wake word, then tune the wake list/threshold.
 
 ## Requirements
 

config.toml

@@ -5,7 +5,7 @@
 # wake phrases for listen mode. fuzzy-matched: case/space-insensitive, lenient on
 # the coined word "claudedo" (whisper renders it inconsistently). number words are
 # normalized to digits before command matching.
-phrases = ["claudedo", "hey claude"]
+phrases = ["claudedo", "claude do", "hey claude", "ok claude", "okay claude"]
 
 [input]
 # "listen" (default): continuous capture; only acts on utterances that start with a
@@ -21,10 +21,10 @@ mode = "listen"
 ptt_key = "space"
 
 [stt]
-# faster-whisper model size. "small" is a good accuracy/latency balance for the
+# faster-whisper model size. "medium" is the default — biggest accuracy gain for the
-# short command grammar (~sub-second per chunk on a strong cpu). if the coined wake
+# coined wake word ("claudedo" / "claude do") and fine on a strong cpu. "small" is
-# word "claudedo" is recognized poorly, bump to "medium" (slower per chunk).
+# faster but less accurate; "large-v3" is most accurate if medium still struggles.
-model = "small"
+model = "medium"
 language = "en"
 # mic device: "auto", or a sounddevice device index (integer) / substring of a
 # device name. run `claudedo test-audio` to list devices.
@@ -36,22 +36,42 @@ compute = "auto"
 # capture parameters. 16 kHz mono is what whisper expects.
 samplerate = 16000
 channels = 1
-# listen-mode silence segmentation: an utterance ends after this many seconds below
+# rms energy below this counts as silence (the VAD onset/endpoint floor).
-# the rms threshold. keeps latency low without streaming.
 silence_threshold = 0.012
-silence_duration = 0.8
 # ignore utterances shorter than this (clicks, coughs).
 min_utterance = 0.3
-# hard cap on a single utterance so a stuck stream can't grow unbounded.
+
-max_utterance = 15.0
+[vad]
+# Alexa-style record-until-pause endpointing (listen mode). capture starts on speech
+# onset and ends after this much trailing silence — the natural end of an utterance.
+# a real pause both ends the command AND separates it from following chatter (the
+# chatter becomes a separate capture that the wake gate then discards).
+silence_ms = 800
+# hard cap so continuous noise can't record forever.
+max_seconds = 10.0
 
 [behavior]
 # dictation never auto-submits: "type <phrase>" inserts literal text only; you say
 # "send" separately to submit (read-before-send).
 type_autosend = false
-# fuzzy match ratio (0..1) required to accept a wake phrase / command token.
+# fuzzy match ratios (0..1). the asymmetry is deliberate: a false WAKE is cheap (it
-match_threshold = 0.8
+# wakes, finds no command, does nothing), so wake is lenient; a false COMMAND fires
+# the WRONG action, so commands stay tight. lower = more lenient = more matches.
+# prefer expanding command synonyms over loosening command_fuzzy_threshold.
+wake_fuzzy_threshold = 0.6
+command_fuzzy_threshold = 0.8
 # optional filler words that may precede a command and are ignored for matching:
 # "select yes" / "use yes" behave like "yes". (a filler word followed by a digit is
 # the select command, e.g. "select 1", and is not dropped.)
 filler_words = ["select", "use", "choose"]
+# when no sticky target is set and exactly ONE claude-* session is running:
+#   false (default) -> require an explicit `set <name>` or one-shot `target <name>`;
+#                      a bare command does nothing and tells you to set one.
+#   true            -> auto-target that single session (convenience).
+auto_target = false
+# DEBUG ONLY — relaxes the privacy invariant. when true, the daemon console prints
+# the raw transcript of EVERY utterance, including non-wake speech it would otherwise
+# drop silently (shown as `heard (dropped): "<transcript>"`). use it to see exactly
+# how Whisper renders your wake word, then turn it OFF. default false: non-wake speech
+# is discarded without ever printing the transcript.
+print_heard = false

install.sh

@@ -94,6 +94,18 @@ mkdir -p "$CONF_DIR"
 install -m 0644 "$REPO_DIR/shell/cc.sh" "$CONF_DIR/cc.sh"
 echo "  wrote $CONF_DIR/cc.sh"
 
+# install config.toml to the standard location so the daemon finds it from any dir.
+# never clobber an edited user config: copy only if absent, else drop a .new to diff.
+if [ ! -f "$CONF_DIR/config.toml" ]; then
+    install -m 0644 "$REPO_DIR/config.toml" "$CONF_DIR/config.toml"
+    echo "  wrote $CONF_DIR/config.toml"
+elif ! cmp -s "$REPO_DIR/config.toml" "$CONF_DIR/config.toml"; then
+    install -m 0644 "$REPO_DIR/config.toml" "$CONF_DIR/config.toml.new"
+    echo "  kept your $CONF_DIR/config.toml; new default written to config.toml.new (diff to merge)"
+else
+    echo "  $CONF_DIR/config.toml already current"
+fi
+
 # wire EVERY rc that exists (the user may have both zsh and bash).
 wired_any=0
 for RC in "$HOME/.zshrc" "$HOME/.bashrc"; do

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "claudedo"
-version = "0.1.1"
+version = "0.1.3"
 description = "voice-control daemon for claude code (local STT -> tmux send-keys)"
 readme = "README.md"
 requires-python = ">=3.10"

src/claudedo/__init__.py

@@ -1,3 +1,3 @@
 """claudedo — voice-control daemon for claude code (local STT -> tmux send-keys)"""
 
-__version__ = "0.1.1"
+__version__ = "0.1.3"

src/claudedo/__main__.py

@@ -33,12 +33,12 @@ def cmd_start(args: argparse.Namespace) -> int:
     config = _load_or_die(args.config)
     if args.mode:
         config.mode = args.mode
-    if not args.skip_audio_check:
+    if args.check:
         print("checking mic before listening (speak briefly) ...")
         peak = _probe_mic(config, seconds=2.0, verbose=False)
         if peak is None or peak < 0.02:
             print("mic check failed — no usable input.", file=sys.stderr)
-            print("run `claudedo test-audio` to debug; or `claudedo start --skip-audio-check`",
+            print("run `claudedo test-audio` to debug, or `claudedo start` to skip the check",
                   file=sys.stderr)
             return 1
         print(f"mic OK (peak {peak:.3f}).")
@@ -217,8 +217,8 @@ def build_parser() -> argparse.ArgumentParser:
 
     sp = sub.add_parser("start", help="run the daemon (foreground)")
     sp.add_argument("--mode", choices=("listen", "ptt"), help="override input mode")
-    sp.add_argument("--skip-audio-check", action="store_true",
+    sp.add_argument("--check", action="store_true",
-                    help="skip the pre-listen mic check")
+                    help="run a mic check before listening (off by default)")
     sp.set_defaults(func=cmd_start)
 
     sub.add_parser("stop", help="stop a running daemon").set_defaults(func=cmd_stop)

src/claudedo/config.py

@@ -44,12 +44,15 @@ class Config:
     samplerate: int
     channels: int
     silence_threshold: float
-    silence_duration: float
+    vad_silence_ms: int
+    vad_max_seconds: float
     min_utterance: float
-    max_utterance: float
     type_autosend: bool
-    match_threshold: float
+    wake_fuzzy_threshold: float
+    command_fuzzy_threshold: float
     filler_words: tuple[str, ...]
+    auto_target: bool
+    print_heard: bool
     source_path: Path | None = field(default=None)
 
 
@@ -96,7 +99,7 @@ def load_config(explicit: str | os.PathLike | None = None) -> Config:
     if mode not in _VALID_MODES:
         raise ConfigError(f"[input].mode must be one of {_VALID_MODES}, got {mode!r}")
 
-    model = _require(raw, "stt", "model", (str,), "small")
+    model = _require(raw, "stt", "model", (str,), "medium")
     if model not in _VALID_MODELS:
         log.warning("unknown stt model %r — passing through to faster-whisper", model)
 
@@ -111,17 +114,25 @@ def load_config(explicit: str | os.PathLike | None = None) -> Config:
         samplerate=int(_require(raw, "audio", "samplerate", (int,), 16000)),
         channels=int(_require(raw, "audio", "channels", (int,), 1)),
         silence_threshold=float(_require(raw, "audio", "silence_threshold", (int, float), 0.012)),
-        silence_duration=float(_require(raw, "audio", "silence_duration", (int, float), 0.8)),
+        vad_silence_ms=int(_require(raw, "vad", "silence_ms", (int,), 800)),
+        vad_max_seconds=float(_require(raw, "vad", "max_seconds", (int, float), 10.0)),
         min_utterance=float(_require(raw, "audio", "min_utterance", (int, float), 0.3)),
-        max_utterance=float(_require(raw, "audio", "max_utterance", (int, float), 15.0)),
         type_autosend=bool(_require(raw, "behavior", "type_autosend", (bool,), False)),
-        match_threshold=float(_require(raw, "behavior", "match_threshold", (int, float), 0.8)),
+        wake_fuzzy_threshold=float(_require(raw, "behavior", "wake_fuzzy_threshold", (int, float), 0.6)),
+        command_fuzzy_threshold=float(_require(raw, "behavior", "command_fuzzy_threshold",
+                                               (int, float), 0.8)),
         filler_words=tuple(_require(raw, "behavior", "filler_words", (list,),
                                     ["select", "use", "choose"])),
+        auto_target=bool(_require(raw, "behavior", "auto_target", (bool,), False)),
+        print_heard=bool(_require(raw, "behavior", "print_heard", (bool,), False)),
         source_path=path,
     )
-    if not 0.0 < cfg.match_threshold <= 1.0:
+    for label, val in (("wake_fuzzy_threshold", cfg.wake_fuzzy_threshold),
-        raise ConfigError("[behavior].match_threshold must be in (0, 1]")
+                       ("command_fuzzy_threshold", cfg.command_fuzzy_threshold)):
+        if not 0.0 < val <= 1.0:
+            raise ConfigError(f"[behavior].{label} must be in (0, 1]")
+    if cfg.vad_silence_ms <= 0 or cfg.vad_max_seconds <= 0:
+        raise ConfigError("[vad].silence_ms and max_seconds must be positive")
     if cfg.samplerate <= 0 or cfg.channels <= 0:
         raise ConfigError("[audio].samplerate and channels must be positive")
     return cfg

src/claudedo/console.py

@@ -0,0 +1,51 @@
+"""colored, prefixed console output for the daemon's recognition/action feed.
+
+every line is ``HH:MM:SS [PREFIX] message``. prefixes group the source: a session
+name (e.g. ``[claude-libs]``) for anything injected into a tmux session, ``[SYSTEM]``
+for daemon-control/state lines, and ``[VOICE]`` for STT/recognition lines. color is
+opt-in via tty detection (or forced): green for successful injections, red for
+drops/errors, dim for routine. falls back to plain text when stdout is not a tty.
+"""
+
+from __future__ import annotations
+
+import sys
+import time
+
+RESET = "\033[0m"
+_COLORS = {
+    "green": "\033[32m",
+    "red": "\033[31m",
+    "yellow": "\033[33m",
+    "cyan": "\033[36m",
+    "dim": "\033[2m",
+    "bold": "\033[1m",
+}
+
+SYSTEM = "SYSTEM"
+VOICE = "VOICE"
+
+
+class Console:
+    """formats and prints daemon log lines with timestamp, prefix, and color"""
+
+    def __init__(self, color: bool | None = None, stream=None, clock=None) -> None:
+        self.stream = stream if stream is not None else sys.stdout
+        self._clock = clock or time.localtime
+        if color is None:
+            color = hasattr(self.stream, "isatty") and self.stream.isatty()
+        self.color = bool(color)
+
+    def _stamp(self) -> str:
+        t = self._clock()
+        return f"{t.tm_hour:02d}:{t.tm_min:02d}:{t.tm_sec:02d}"
+
+    def _paint(self, text: str, color: str | None) -> str:
+        if not self.color or not color or color not in _COLORS:
+            return text
+        return f"{_COLORS[color]}{text}{RESET}"
+
+    def emit(self, prefix: str, message: str, color: str | None = None) -> None:
+        """print one line: ``HH:MM:SS [prefix] message`` (message optionally colored)"""
+        line = f"{self._stamp()} {self._paint(f'[{prefix}]', 'dim')} {self._paint(message, color)}"
+        print(line, file=self.stream, flush=True)

src/claudedo/daemon.py

@@ -18,6 +18,7 @@ from pathlib import Path
 
 from . import audio, grammar, inject, target
 from .config import Config
+from .console import SYSTEM, VOICE, Console
 from .stt import Transcriber
 
 log = logging.getLogger(__name__)
@@ -114,6 +115,8 @@ class Daemon:
         self._transcriber: Transcriber | None = None
         self._device: int | None = None
         self._ptt = _PTTKey()
+        self._pending: dict[str, int] = {}
+        self._console = Console()
 
     def _install_signals(self) -> None:
         signal.signal(signal.SIGTERM, self._on_signal)
@@ -133,6 +136,7 @@ class Daemon:
             model=cfg.stt_model, language=cfg.stt_language,
             device=cfg.stt_compute if cfg.stt_compute in ("cpu", "cuda") else "auto",
             compute_type="auto",
+            initial_prompt=grammar.initial_prompt(cfg.wake_phrases),
         )
         if audio.warm_up(cfg.samplerate, cfg.channels, self._device):
             log.info("mic warmed up (source live)")
@@ -148,22 +152,22 @@ class Daemon:
             return audio.record_while(
                 cfg.samplerate, cfg.channels, self._device,
                 held=lambda: not self._ptt.wait_press(self.stopped),
-                max_utterance=cfg.max_utterance, min_utterance=cfg.min_utterance,
+                max_utterance=cfg.vad_max_seconds, min_utterance=cfg.min_utterance,
             )
         return audio.record_until_silence(
             cfg.samplerate, cfg.channels, self._device,
-            silence_threshold=cfg.silence_threshold, silence_duration=cfg.silence_duration,
+            silence_threshold=cfg.silence_threshold, silence_duration=cfg.vad_silence_ms / 1000.0,
-            min_utterance=cfg.min_utterance, max_utterance=cfg.max_utterance,
+            min_utterance=cfg.min_utterance, max_utterance=cfg.vad_max_seconds,
             stop=self.stopped,
         )
 
     def _handle(self, transcript: str) -> None:
         cfg = self.config
         require_wake = self.mode == "listen"
-        parsed = grammar.parse(transcript, cfg.wake_phrases, cfg.match_threshold, require_wake,
+        parsed = grammar.parse(transcript, cfg.wake_phrases, cfg.wake_fuzzy_threshold,
-                               filler=cfg.filler_words)
+                               cfg.command_fuzzy_threshold, require_wake, filler=cfg.filler_words)
         if parsed is None or parsed.action is None:
-            self._emit(f'heard: "{transcript}" -> no command matched')
+            self._console.emit(VOICE, f'heard "{transcript}" -> no command matched', "yellow")
             return
         action = parsed.action
 
@@ -171,36 +175,78 @@ class Daemon:
             new_mode = str(action.arg)
             if new_mode != self.mode:
                 self.mode = new_mode
-                self._emit(f"mode -> {new_mode}")
+                self._console.emit(SYSTEM, f"mode -> {new_mode}", "cyan")
                 self._refresh_state()
             return
         if action.name == "set":
             session = target.set_target(str(action.arg))
-            self._emit(f"set sticky -> {session}")
+            self._pending.pop(session, None)
+            self._console.emit(SYSTEM, f"set sticky -> {session}", "cyan")
             self._refresh_state()
             return
         if action.name == "unset":
             target.unset_target()
-            self._emit("unset (cleared)")
+            self._console.emit(SYSTEM, "unset (cleared)", "cyan")
             self._refresh_state()
             return
         if action.name == "list":
             sessions = target.list_sessions()
-            self._emit("list -> " + (", ".join(sessions) if sessions else "(none running)"))
+            self._console.emit(SYSTEM, "list -> " + (", ".join(sessions) if sessions else "(none running)"))
+            return
+        if action.name == "debug":
+            self._console.emit(VOICE, f'debug: "{action.arg}"', "yellow")
             return
 
-        session, reason = target.resolve(parsed.one_shot)
+        session, reason = target.resolve(parsed.one_shot, auto_target=cfg.auto_target)
         if session is None:
-            self._emit(f'heard: "{transcript}" -> {reason} -> matched {self._describe(action)} '
+            self._console.emit(VOICE, f'heard "{transcript}" -> {reason} -> '
-                       f'-> did nothing')
+                               f'{self._describe(action)} did nothing', "red")
             return
-        prefix = f'heard: "{transcript}" -> {reason} -> matched {self._describe(action)}'
+        self._inject(session, transcript, reason, action)
-        if action.name == "type" and not cfg.type_autosend:
+
-            inject.send_literal(session, str(action.arg))
+    def _inject(self, session: str, transcript: str, reason: str, action) -> None:
-            self._emit(f"{prefix} -> injected literal {str(action.arg)!r} -> {session}")
+        """run a resolved command against `session`, tracking the uncommitted-input
+        buffer so backspace/erase delete only back to the last submit boundary"""
+        heard = f'heard "{transcript}" ({reason})'
+        name = action.name
+
+        if name == "type":
+            text = str(action.arg)
+            inject.send_literal(session, text)
+            self._pending[session] = self._pending.get(session, 0) + len(text)
+            if self.config.type_autosend:
+                inject.send_named(session, inject.keys.SUBMIT)
+                self._pending[session] = 0
+            self._console.emit(session, f"{heard} -> typed {text!r}"
+                               + (" + send" if self.config.type_autosend else ""), "green")
             return
+        if name == "space":
+            n = int(action.arg)
             inject.perform(session, action)
-        self._emit(f"{prefix} -> injected {self._describe(action)} -> {session}")
+            self._pending[session] = self._pending.get(session, 0) + n
+            self._console.emit(session, f"{heard} -> space x{n}", "green")
+            return
+        if name == "backspace":
+            have = self._pending.get(session, 0)
+            n = min(int(action.arg), have)
+            if n:
+                inject.perform(session, grammar.Action("backspace", n))
+                self._pending[session] = have - n
+            self._console.emit(session, f"{heard} -> backspace x{n}"
+                               + ("" if n == int(action.arg) else " (capped at boundary)"), "green")
+            return
+        if name == "erase":
+            n = self._pending.get(session, 0)
+            if n:
+                inject.perform(session, grammar.Action("erase", n))
+                self._pending[session] = 0
+            self._console.emit(session, f"{heard} -> erase x{n} (to last boundary)", "green")
+            return
+
+        inject.perform(session, action)
+        if name == "submit":
+            self._pending[session] = 0
+        self._console.emit(session, f"{heard} -> {self._describe(action)}", "green")
 
     @staticmethod
     def _describe(action) -> str:
@@ -208,11 +254,6 @@ class Daemon:
             return action.name.upper()
         return f"{action.name.upper()}({action.arg})"
 
-    @staticmethod
-    def _emit(line: str) -> None:
-        """print a recognition/action line to the watched terminal"""
-        print(line, flush=True)
-
     def _has_wake(self, transcript: str) -> bool:
         """true if the utterance starts with a wake phrase (listen-mode gate).
 
@@ -220,20 +261,17 @@ class Daemon:
         invariant: non-command speech is discarded, never recorded.
         """
         cfg = self.config
-        return grammar.strip_wake(transcript, cfg.wake_phrases, cfg.match_threshold, True) is not None
+        return grammar.strip_wake(transcript, cfg.wake_phrases,
+                                  cfg.wake_fuzzy_threshold, True) is not None
 
     def _print_startup(self) -> None:
         cfg = self.config
         dev = cfg.stt_device if cfg.stt_device != "auto" else "default"
-        target_now = target.read_active() or "(none — run cc to attach)"
+        target_now = target.read_active() or "(none — run cc / set <name>)"
-        self._emit("── claudedo ─────────────────────────────────")
+        self._console.emit(SYSTEM, f"claudedo {self.mode} mode — Ctrl-C to stop", "bold")
-        self._emit(f"  model:   {cfg.stt_model} ({cfg.stt_language})")
+        self._console.emit(SYSTEM, f"model {cfg.stt_model} ({cfg.stt_language}) · mic {dev} · "
-        self._emit(f"  mic:     {dev}")
+                           f"target {target_now}")
-        self._emit(f"  mode:    {self.mode}")
+        self._console.emit(SYSTEM, "wake: " + ", ".join(cfg.wake_phrases))
-        self._emit(f"  target:  {target_now}")
-        self._emit(f"  wake:    {', '.join(cfg.wake_phrases)}")
-        self._emit("  Ctrl-C to stop")
-        self._emit("─────────────────────────────────────────────")
 
     def _refresh_state(self) -> None:
         write_state(os.getpid(), self.mode, target.read_active())
@@ -257,7 +295,10 @@ class Daemon:
                 if not transcript:
                     continue
                 if self.mode == "listen" and not self._has_wake(transcript):
-                    self._emit("dropped: non-wake speech (not recorded)")
+                    if self.config.print_heard:
+                        self._console.emit(VOICE, f'heard (dropped) "{transcript}"', "red")
+                    else:
+                        self._console.emit(VOICE, "dropped: non-wake speech (not recorded)", "dim")
                     continue
                 self._handle(transcript)
         finally:

src/claudedo/grammar.py

@@ -27,11 +27,39 @@ _NUMBER_WORDS = {
 
 _INDEX_WORDS = {"1": 1, "2": 2, "3": 3, "4": 4}
 
+_COUNT_WORDS = {
+    "five": 5, "six": 6, "seven": 7, "eight": 8, "nine": 9, "ten": 10,
+    "eleven": 11, "twelve": 12, "thirteen": 13, "fourteen": 14, "fifteen": 15,
+    "sixteen": 16, "seventeen": 17, "eighteen": 18, "nineteen": 19, "twenty": 20,
+}
+
+_YES_VERBS = ("yes", "yeah", "yep", "yup")
+_NO_VERBS = ("no", "nope", "nah")
+_APPROVE_VERBS = ("approve", "allow")
+_DENY_VERBS = ("deny", "reject")
+_SUBMIT_VERBS = ("send", "enter", "submit")
+_CANCEL_VERBS = ("cancel", "escape")
+_TYPE_VERBS = ("type", "dictate", "write")
+_BACKSPACE_VERBS = ("backspace", "delete")
+_SPACE_VERBS = ("space", "spacebar")
+_ADD_VERBS = ("add", "insert")
+_ERASE_VERBS = ("erase", "clear", "wipe")
+_DEBUG_VERBS = ("debug", "echo")
+_MODE_VERBS = ("mode",)
 _STICKY_VERBS = ("set", "sticky", "switch")
 _ONESHOT_VERBS = ("target",)
 _UNSET_VERBS = ("unset", "unsticky")
 _LIST_VERBS = ("list", "sessions")
 _SELECT_VERBS = ("select", "option", "choose", "number")
+
+# every command/synonym word, for biasing the STT toward the vocabulary we expect.
+_COMMAND_WORDS = (
+    _YES_VERBS + _NO_VERBS + _APPROVE_VERBS + _DENY_VERBS + _SUBMIT_VERBS
+    + _CANCEL_VERBS + _TYPE_VERBS + _BACKSPACE_VERBS + _SPACE_VERBS + _ADD_VERBS
+    + _ERASE_VERBS + _DEBUG_VERBS + _MODE_VERBS + _STICKY_VERBS + _ONESHOT_VERBS + _UNSET_VERBS
+    + _LIST_VERBS + _SELECT_VERBS + ("ptt", "listen")
+    + ("one", "two", "three", "four")
+)
 DEFAULT_FILLER = ("select", "use", "choose")
 
 
@@ -39,9 +67,10 @@ DEFAULT_FILLER = ("select", "use", "choose")
 class Action:
     """a matched command: a name plus an optional argument.
 
-    names: yes, no, select, approve, deny, submit, type, cancel, mode, set, unset,
+    names: yes, no, select, approve, deny, submit, type, space, backspace, erase,
-    list. arg carries the select index (int), the literal text for ``type``, the mode
+    cancel, mode, set, unset, list. arg carries the select index (int), the literal
-    for ``mode``, or the session short-name for ``set``.
+    text for ``type``, the count for ``space``/``backspace`` (int), the mode for
+    ``mode``, or the session short-name for ``set``.
     """
 
     name: str
@@ -72,6 +101,26 @@ def normalize(text: str) -> str:
     return " ".join(tokens)
 
 
+def vocabulary(wake_phrases: list[str]) -> list[str]:
+    """the wake + command vocabulary, deduped in first-seen order.
+
+    single source for biasing the STT: the same wake phrases the matcher uses plus
+    every command/synonym word in _COMMAND_WORDS. no separate hardcoded copy.
+    """
+    seen: dict[str, None] = {}
+    for word in list(wake_phrases) + list(_COMMAND_WORDS):
+        key = word.strip()
+        if key and key not in seen:
+            seen[key] = None
+    return list(seen)
+
+
+def initial_prompt(wake_phrases: list[str]) -> str:
+    """a comma-joined vocabulary string to pass faster-whisper as initial_prompt,
+    conditioning transcription toward the words we expect (esp. the coined wake)"""
+    return ", ".join(vocabulary(wake_phrases))
+
+
 def _ratio(a: str, b: str) -> float:
     return SequenceMatcher(None, a, b).ratio()
 
@@ -123,6 +172,22 @@ def _fuzzy_in(token: str, options: tuple[str, ...], threshold: float) -> bool:
     return any(_ratio(token, opt) >= threshold for opt in options)
 
 
+def _leading_count(rest: list[str], default: int = 1) -> int:
+    """read a count from the first token (digit or number word), else the default.
+
+    'backspace 3' -> 3, 'backspace ten' -> 10 (normalize maps small words to digits;
+    larger words come from _COUNT_WORDS), 'backspace' -> default.
+    """
+    if not rest:
+        return default
+    tok = rest[0]
+    if tok.isdigit():
+        return max(0, int(tok))
+    if tok in _COUNT_WORDS:
+        return _COUNT_WORDS[tok]
+    return default
+
+
 def match_command(remainder: str, threshold: float) -> Action | None:
     """map a normalized command remainder to an Action, or None if unrecognized.
 
@@ -140,27 +205,42 @@ def match_command(remainder: str, threshold: float) -> Action | None:
     if head in _INDEX_WORDS:
         return Action("select", _INDEX_WORDS[head])
 
-    if _fuzzy_in(head, ("yes", "yeah", "yep", "yup"), threshold):
+    if _fuzzy_in(head, _YES_VERBS, threshold):
         return Action("yes")
-    if _fuzzy_in(head, ("no", "nope", "nah"), threshold):
+    if _fuzzy_in(head, _NO_VERBS, threshold):
         return Action("no")
-    if _fuzzy_in(head, ("approve", "allow"), threshold):
+    if _fuzzy_in(head, _APPROVE_VERBS, threshold):
         return Action("approve")
-    if _fuzzy_in(head, ("deny", "reject"), threshold):
+    if _fuzzy_in(head, _DENY_VERBS, threshold):
         return Action("deny")
-    if _fuzzy_in(head, ("send", "enter", "submit"), threshold):
+    if _fuzzy_in(head, _SUBMIT_VERBS, threshold):
         return Action("submit")
-    if _fuzzy_in(head, ("cancel", "escape"), threshold):
+    if _fuzzy_in(head, _CANCEL_VERBS, threshold):
         return Action("cancel")
 
     if _fuzzy_in(head, _SELECT_VERBS, threshold) and rest and rest[0] in _INDEX_WORDS:
         return Action("select", _INDEX_WORDS[rest[0]])
 
-    if _fuzzy_in(head, ("type", "dictate", "write"), threshold):
+    if _fuzzy_in(head, _TYPE_VERBS, threshold):
         text = " ".join(rest).strip()
         return Action("type", text) if text else None
 
-    if _fuzzy_in(head, ("mode",), threshold) and rest:
+    if _fuzzy_in(head, _BACKSPACE_VERBS, threshold):
+        return Action("backspace", _leading_count(rest, default=1))
+    if _fuzzy_in(head, _SPACE_VERBS, threshold):
+        return Action("space", _leading_count(rest, default=1))
+    if _fuzzy_in(head, _ADD_VERBS, threshold) and rest:
+        tail = [t for t in rest if t not in ("a", "an")]
+        if any(_fuzzy_in(t, ("space", "spaces"), threshold) for t in tail):
+            count = next((int(t) for t in tail if t.isdigit()),
+                         next((_COUNT_WORDS[t] for t in tail if t in _COUNT_WORDS), 1))
+            return Action("space", count)
+    if _fuzzy_in(head, _ERASE_VERBS, threshold):
+        return Action("erase")
+    if _fuzzy_in(head, _DEBUG_VERBS, threshold):
+        return Action("debug", " ".join(rest).strip())
+
+    if _fuzzy_in(head, _MODE_VERBS, threshold) and rest:
         if _fuzzy_in(rest[0], ("ptt",), threshold) or "push" in rest[0]:
             return Action("mode", "ptt")
         if _fuzzy_in(rest[0], ("listen",), threshold):
@@ -191,24 +271,28 @@ def _strip_filler(tokens: list[str], filler: tuple[str, ...], threshold: float)
     return tokens
 
 
-def parse(transcript: str, wake_phrases: list[str], threshold: float,
+def parse(transcript: str, wake_phrases: list[str], wake_threshold: float,
-          require_wake: bool, filler: tuple[str, ...] = DEFAULT_FILLER) -> ParsedCommand | None:
+          command_threshold: float, require_wake: bool,
+          filler: tuple[str, ...] = DEFAULT_FILLER) -> ParsedCommand | None:
     """full parse: wake gate -> optional one-shot target -> filler -> command.
 
-    returns a ParsedCommand (one_shot, action), or None if the wake gate dropped the
+    wake_threshold gates the wake phrase (lenient — a false wake is cheap, it just
-    utterance (listen mode, no wake phrase). a ParsedCommand with action=None means a
+    finds no command); command_threshold gates the command words (stricter — a false
-    wake phrase was present but no command matched.
+    command fires the wrong action). returns a ParsedCommand (one_shot, action), or
+    None if the wake gate dropped the utterance (listen mode, no wake phrase). a
+    ParsedCommand with action=None means a wake phrase was present but no command
+    matched.
     """
-    remainder = strip_wake(transcript, wake_phrases, threshold, require_wake)
+    remainder = strip_wake(transcript, wake_phrases, wake_threshold, require_wake)
     if remainder is None:
         return None
 
     tokens = remainder.split(" ") if remainder else []
     one_shot: str | None = None
-    if tokens and _fuzzy_in(tokens[0], _ONESHOT_VERBS, threshold) and len(tokens) >= 2:
+    if tokens and _fuzzy_in(tokens[0], _ONESHOT_VERBS, command_threshold) and len(tokens) >= 2:
         one_shot = tokens[1]
         tokens = tokens[2:]
 
-    tokens = _strip_filler(tokens, filler, threshold)
+    tokens = _strip_filler(tokens, filler, command_threshold)
-    action = match_command(" ".join(tokens), threshold)
+    action = match_command(" ".join(tokens), command_threshold)
     return ParsedCommand(one_shot=one_shot, action=action)

src/claudedo/inject.py

@@ -45,11 +45,18 @@ class OutputHandler(ABC):
     def send_literal(self, session: str, text: str) -> None:
         """emit literal text into the input box without submitting (``type``)"""
 
+    def send_repeat(self, session: str, token: str, count: int) -> None:
+        """emit a named key `count` times (e.g. BSpace x n). default impl loops."""
+        if count <= 0:
+            return
+        self.send_named(session, [token] * count)
+
     def perform(self, session: str, action) -> bool:
         """resolve a grammar.Action to keystrokes and emit them. returns acted?.
 
-        ``switch`` and ``mode`` are handled by the daemon (they change daemon state,
+        ``switch``/``set``/``mode`` etc. are handled by the daemon (they change daemon
-        not the claude session), so they are ignored here.
+        state, not the claude session), so they are ignored here. ``erase`` arrives
+        with action.arg already set to the count the daemon wants backspaced.
         """
         name = action.name
         if name == "yes":
@@ -72,6 +79,10 @@ class OutputHandler(ABC):
             self.send_named(session, seq)
         elif name == "type":
             self.send_literal(session, str(action.arg))
+        elif name == "space":
+            self.send_literal(session, " " * int(action.arg))
+        elif name in ("backspace", "erase"):
+            self.send_repeat(session, keys.BACKSPACE[0], int(action.arg))
         else:
             return False
         return True

src/claudedo/keys.py

@@ -37,6 +37,12 @@ DENY = ["3"]
 SUBMIT = ["Enter"]
 CANCEL = ["Escape"]
 
+# BACKSPACE deletes one char left; SPACE inserts one literal space. both are emitted
+# repeatedly for `backspace <n>` / `space <n>` and for `erase` (n = the daemon's
+# tracked uncommitted-input count). BSpace is tmux's name for the backspace key.
+BACKSPACE = ["BSpace"]
+SPACE = [" "]
+
 SELECT_BY_INDEX = {
     1: SELECT_1,
     2: SELECT_2,

src/claudedo/stt.py

@@ -6,32 +6,106 @@ short in-memory chunk; nothing is written to disk or sent anywhere.
 
 from __future__ import annotations
 
+import contextlib
 import logging
+import os
+import re
+import sys
 
 import numpy as np
 
 log = logging.getLogger(__name__)
 
+_NOISE = re.compile(r"GPU device discovery failed|device_discovery\.cc|DiscoverDevicesForPlatform")
+
+
+def _quiet_backends() -> None:
+    """quiet onnxruntime/ctranslate2 chatter and the faster_whisper INFO log.
+
+    faster-whisper's VAD loads an onnx model whose device discovery prints a noisy
+    'GPU device discovery failed' warning on headless/WSL hosts with no GPU sysfs.
+    the env var + logger severity stop most onnx logging; the warning itself is
+    emitted at C++ init and is filtered out of stderr by _filter_stderr().
+    """
+    os.environ.setdefault("ORT_LOGGING_LEVEL", "3")
+    os.environ.setdefault("OMP_NUM_THREADS", os.environ.get("OMP_NUM_THREADS", "4"))
+    logging.getLogger("faster_whisper").setLevel(logging.WARNING)
+    try:
+        import onnxruntime
+        onnxruntime.set_default_logger_severity(3)
+    except Exception:
+        pass
+
+
+@contextlib.contextmanager
+def _filter_stderr():
+    """drop onnxruntime's GPU-discovery warning lines from stderr for this block.
+
+    a pipe temporarily replaces fd 2; a pump thread forwards every line to the real
+    stderr EXCEPT the known GPU-discovery noise, so real errors still surface. the
+    original fd is always restored on exit.
+    """
+    import threading
+
+    try:
+        stderr_fd = sys.stderr.fileno()
+    except (AttributeError, OSError):
+        yield
+        return
+
+    saved_fd = os.dup(stderr_fd)
+    read_fd, write_fd = os.pipe()
+    os.dup2(write_fd, stderr_fd)
+    os.close(write_fd)
+
+    def pump():
+        with os.fdopen(read_fd, "rb") as reader, os.fdopen(saved_fd, "wb", closefd=False) as out:
+            for line in reader:
+                if not _NOISE.search(line.decode("utf-8", "replace")):
+                    out.write(line)
+                    out.flush()
+
+    thread = threading.Thread(target=pump, daemon=True)
+    thread.start()
+    try:
+        yield
+    finally:
+        import time
+
+        time.sleep(0.05)
+        os.dup2(saved_fd, stderr_fd)
+        os.close(saved_fd)
+        thread.join(timeout=1.0)
+
 
 class Transcriber:
     """a loaded faster-whisper model that transcribes float32 mono audio chunks"""
 
     def __init__(self, model: str = "small", language: str = "en", device: str = "auto",
-                 compute_type: str = "auto") -> None:
+                 compute_type: str = "auto", initial_prompt: str | None = None) -> None:
         self.language = language
+        self.initial_prompt = initial_prompt
         self._model = self._load(model, device, compute_type)
+        self._warm()
 
     @staticmethod
     def _load(model: str, device: str, compute_type: str):
-        from faster_whisper import WhisperModel
-
         if device == "auto":
             device = "cpu"
         if compute_type == "auto":
             compute_type = "int8" if device == "cpu" else "float16"
         log.info("loading faster-whisper model=%s device=%s compute=%s", model, device, compute_type)
+        with _filter_stderr():
+            _quiet_backends()
+            from faster_whisper import WhisperModel
             return WhisperModel(model, device=device, compute_type=compute_type)
 
+    def _warm(self) -> None:
+        """run one throwaway transcribe so the VAD onnx session inits now, under the
+        stderr filter — the GPU-discovery warning fires here once, not in the loop"""
+        with _filter_stderr():
+            list(self._model.transcribe(np.zeros(1600, dtype=np.float32), vad_filter=True)[0])
+
     def transcribe(self, audio: np.ndarray, samplerate: int = 16000) -> str:
         """transcribe a mono float32 numpy array to a stripped text string.
 
@@ -47,6 +121,7 @@ class Transcriber:
             beam_size=1,
             vad_filter=True,
             condition_on_previous_text=False,
+            initial_prompt=self.initial_prompt,
         )
         text = " ".join(seg.text for seg in segments).strip()
         return text

src/claudedo/target.py

@@ -85,7 +85,7 @@ def list_sessions() -> list[str]:
     return sorted(n for n in names if n.startswith(SESSION_PREFIX))
 
 
-def resolve(one_shot: str | None = None) -> tuple[str | None, str]:
+def resolve(one_shot: str | None = None, auto_target: bool = False) -> tuple[str | None, str]:
     """resolve the destination session and a short reason describing the choice.
 
     single source of truth for targeting, used by both the voice and CLI paths.
@@ -95,7 +95,9 @@ def resolve(one_shot: str | None = None) -> tuple[str | None, str]:
     1. one-shot present -> claude-<name> for THIS command only; never falls through
        to a different session if it doesn't exist (explicit beats convenience).
     2. sticky set + exists -> use it.
-    3. nothing sticky, exactly one claude-* session -> auto-use it.
+    3. nothing sticky, exactly one claude-* session:
+         auto_target=True  -> auto-use it;
+         auto_target=False -> require an explicit set/target, do nothing.
     4. nothing sticky, multiple sessions -> ambiguous, do nothing.
     5. nothing sticky, zero sessions -> do nothing.
     """
@@ -113,7 +115,9 @@ def resolve(one_shot: str | None = None) -> tuple[str | None, str]:
 
     sessions = list_sessions()
     if len(sessions) == 1:
+        if auto_target:
             return sessions[0], f"auto-target {sessions[0]} (only session)"
+        return None, f"no target set ({sessions[0]} running — set one)"
     if len(sessions) > 1:
         return None, f"no target set, {len(sessions)} sessions (set one)"
     return None, "no claude sessions"