docs: document voice trigger configuration

docs/USAGE.md

@@ -1706,6 +1706,29 @@ TTS behavior:
 - TTS shares the audio output with music playback
 - Text is limited to 500 characters
 
+### Always-On Trigger Mode
+
+Set a trigger word to enable always-on voice listening. The bot
+continuously transcribes voice and watches for the trigger word. When
+detected, the text after the trigger is spoken back via TTS. No
+`!listen` command needed.
+
+```toml
+[voice]
+trigger = "claude"
+```
+
+Behavior:
+
+- Listener starts automatically on connect (no `!listen on` required)
+- All speech is transcribed and checked for the trigger word
+- Trigger match is case-insensitive: "Claude", "CLAUDE", "claude" all work
+- On match, the trigger word is stripped and the remainder is sent to TTS
+- Non-triggered speech is silently discarded (unless `!listen` is also on)
+- When both trigger and `!listen` are active, triggered speech goes to
+  TTS and all other speech is posted as the usual "heard X say: ..."
+- `!listen` status shows trigger configuration when set
+
 Configuration (optional):
 
 ```toml
@@ -1713,4 +1736,5 @@ Configuration (optional):
 whisper_url = "http://192.168.122.1:8080/inference"
 piper_url = "http://192.168.122.1:5000/"
 silence_gap = 1.5
+trigger = ""
 ```