Adom Core

Make voice a first-class, pre-installed capability (Adom TTS + Hands-Free)

The failure (real, 2026-08-31): a user driving asked an AI to "use the hands free skill." Neither adom/adom-tts nor adom/hands-free was installed in the container, so the AI searched only installed skills, found nothing, and fell back to raw Windows SAPI TTS — a genuinely bad voice and experience. Both packages exist on the wiki and install and work perfectly; they were simply never on the machine.

Root cause — voice is not in the dependency tree. adom/core (v4.13.14) has 10 members: adom, adom-cli, adom-parts-search, adom-vscode, adom-wiki-cli, building-adom-apps, definitions, hook, ralph-loop-test, wiki. Neither adom-tts nor hands-free is a member, and neither is a dependency of adom/hydrogen-bootstrap. So no container — Hydrogen on any platform — ships with a voice. Discovery-by-trigger can't save it: an uninstalled skill's triggers aren't loaded.

Contributing cause — adom/adom-tts's package.json discovery_triggers is empty. The app that actually speaks is undiscoverable in wiki search by phrase; only the separate hands-free skill carries the "I'm driving / read it to me" triggers (and its SKILL.md is excellent). If hands-free isn't installed, nothing catches the intent.

The plan

1. Add adom/adom-tts (and adom/hands-free) as members of adom/core. Voice is a platform-wide capability exactly like adom-parts-search already is — it belongs in core so EVERY bootstrap (Hydrogen Windows/Mac/Ubuntu and any org that builds on core) inherits it. hands-free depends on adom-tts, so adding hands-free pulls both. Bump core; pkg install adom/core re-resolves on every container via the hook auto-update, so the fleet gets a voice with no per-user action. (This is the "auto-install Adom TTS into Hydrogen" the user asked for, done at the right layer so it reaches all platforms, not just Hydrogen.)

2. Populate adom/adom-tts discovery_triggers to match its SKILL.md: "text to speech", "read it to me", "read aloud", "say it out loud", "narrate", "voiceover", "drive mode", "I'm driving", "talk to me instead of writing", etc. So the app surfaces in wiki discovery for anyone browsing, not only via the hands-free skill.

3. Behavioral rule for every Adom AI: when a user names a skill/app that isn't in the installed set, adom-wiki discover search the WIKI before falling back to any lesser mechanism — and offer to install it. Document this in a core skill (definitions or adom) so it's a fleet-wide reflex, not a per-agent accident. This is the specific lesson from the SAPI fallback: the capability existed one pkg install away.

4. (Optional) Dock card / featured for the TTS player, like parts-search, so voice is visibly discoverable in the dock too.

Net: after (1)+(3), a brand-new user who says "read it to me while I drive" gets the real Andrew voice through the shared player on the first try, on any platform — no bad-SAPI experience possible.

Filed from the altium thread after hitting this live.