Hands-Free — voice narration

@@ -1,61 +1,55 @@ ----name: im-driving+name: hands-free+user-invocable: true description: >-  Hands-free / drive-mode. The MOMENT the user signals they're driving or can't-  look at the screen, STOP writing text walls — narrate your answer ALOUD through-  the Adom Desktop TTS player (adom-tts). Read + apply this skill immediately on-  any hands-free phrase, and keep narrating every reply until they say they're done.-  TRIGGERS (invoke instantly on ANY of these, exact or paraphrased):-  "I'm driving", "I am driving", "while I drive", "while driving", "I'm in the car",-  "on the road", "hands free", "hands-free", "drive mode", "driving now",-  "read me", "read it to me", "read that back", "read me your last",-  "read me that", "read me the last", "read your last answer", "speak that",-  "say that out loud", "play it back", "play that", "narrate this", "talk to me",-  "I can't look at the screen", "can't read the screen", "can't see the screen".-metadata:-  type: reference+  Hands-free / eyes-free narration. The MOMENT the user signals they can't look+  at the screen — or just prefers to LISTEN — stop writing text walls and speak+  your answer aloud through the shared Adom TTS player, and keep narrating+  every reply until they say they're done. TRIGGERS (invoke instantly on ANY of+  these, exact or paraphrased): "hands free", "hands-free", "read it to me",+  "read me your answer", "read me that", "read that back", "read your last+  answer", "speak that", "say that out loud", "narrate this", "talk to me",+  "play it back", "I can't look at the screen", "can't read the screen",+  "can't see the screen", "I'm driving", "I am driving", "while I drive",+  "I'm in the car", "on the road", "on my commute", "drive mode". --- -# I'm Driving — hands-free narration+# Hands-Free — voice narration -When the user is driving, **your reply is AUDIO, not text.** Narrate it through-`adom-tts` (the Adom Desktop TTS player). Don't make them read.+The user's eyes and hands are busy (or they simply prefer listening). Speak+your answers; don't write walls of text they can't read. -## ⚠️ The one rule that makes it actually audible (the usual failure)+## The one pattern (everything else lives in the adom-tts skill) -`adom-tts` defaults to a **hydrogen webview** player. If the user's window mode is-**pup**, that webview tab isn't open and **they hear NOTHING** — the clip-synthesizes, you think it played, and they get silence. So ALWAYS match the play-**surface** to their window mode:+This skill is a thin entry point. The canonical contract — verified playback,+the shared queue, History and replays, buffering through connection drops —+is documented ONCE, in the `adom-tts` skill. Read it. The short version:  ```bash-# 1. Detect the surface from the user's window mode-MODE=$(adom-cli hydrogen probe | python3 -c "import sys,json;print(json.load(sys.stdin).get('window_mode','hydrogen'))")-SURFACE=$([ "$MODE" = "pup" ] && echo pup || echo hydrogen)--# 2. If the relay has >1 client, the pup player's browser_open_window is-#    AMBIGUOUS and fails silently. Pin it to the user's host:-export ADOM_DESKTOP_TARGET=$(adom-desktop status | python3 -c "import sys,json;print(json.load(sys.stdin).get('hostname',''))")--# 3. Narrate to THAT surface. NOTE: --thread is a `play` flag, NOT a `say` flag. adom-tts say "This is the <thread> thread — <your spoken answer>" \-  --play --surface "$SURFACE" --out /tmp/reply.mp3+  --out /tmp/answer.mp3 --thread "<thread>" \+  --desc "<one human sentence: what this update says>" \+  --play --confirm ``` -## How to narrate well+- `--thread` + `--desc` are REQUIRED with `--play` (exit 6 otherwise) — many+  AI threads share one player and one History; the human needs WHO and WHAT.+- `--confirm` blocks until the human actually heard it. On UNCONFIRMED, do+  NOT re-send (clips buffer + retry on their own); poll `adom-tts check`.+- Keep each spoken reply conversational, roughly 30-60 seconds.+- No separate text summary on top — the user isn't reading.++## Session rules -- **Start with** "This is the `<thread name>` thread —" (all Claude threads share-  one player, so the user needs to know who's speaking). Also pass `--thread`.-- **Conversational + short.** No markdown, no code, no slugs/jargon read aloud.-- **Proactive.** While they're driving, narrate EVERY reply automatically — don't-  wait to be asked again.-- **Replay:** `adom-tts play last --surface "$SURFACE"`.-- **Confirm if it matters:** add `--wait` (blocks until the audio's `ended` event).-  If it times out, the surface was wrong — re-check `window_mode`.+- Once triggered, narrate EVERY reply until the user ends it ("I'm not+  driving anymore", "back at my desk", "you can stop reading now").+- If the player surface is a desktop browser context (`adom-cli hydrogen+  probe` says window_mode "pup"), add `--surface pup`.+- Missed clips: the player's History drawer replays any clip instantly —+  tell the user it's there if they say they missed something. -## Why this skill exists+## Safety note -The hands-free capability lives in the `adom-tts` tool, but it was undiscoverable-(the `driving-*` skills are about driving *software*, not a car) and it silently-failed when the play surface didn't match the user's window mode. This skill is the-discoverable entry point + the surface-matching fix.+This skill exists so the user does NOT have to look at a screen. Never ask+them to read, click, or review anything visual while hands-free — offer to+queue visual items for later instead.