/setup

Walk the user through configuring the auto-speak plugin on this machine. Be conversational and do one step at a time. The result is a config file at ~/.claude/auto-speak.conf.

IMPORTANT: do not use emojis anywhere in your responses during this walkthrough (or in any test text you pass to say) - your replies are being spoken aloud, and the speech synthesizer pronounces emojis literally ("party popper", "rocket", etc.).

1. Check dependencies

• Run which jq. If missing, tell the user to run brew install jq (the hook is silent without it) and wait for confirmation.
• Confirm say works: say "Setup is starting" so the user hears audio output is functional.

2. Pick a voice

• List good installed voices: say -v '?' | grep -iE "premium|enhanced" and also show a few standard en_US voices if none are enhanced.
• If no Enhanced/Premium voices are installed, recommend downloading one: System Settings > Accessibility > Spoken Content > System Voice > Manage Voices. If that panel freezes (known macOS bug), use VoiceOver Utility > Speech > Voice dropdown > Customize instead. Warn that SIRI VOICES DO NOT WORK with say - do not download those.
• Audition candidates the user is interested in, one at a time: say -v "<Voice Name>" "This is what I sound like reading your responses."
• Let the user pick. An empty voice means the system default.

3. Pick a rate

Demo a few speeds with the chosen voice, e.g. 175 (macOS default), 210, 230: say -v "<Voice>" -r 210 "This is two hundred ten words per minute."

4. Pick a mode

Explain the four options:

• sentence - speaks just the first sentence (a spoken headline)
• paragraph - speaks up to the first blank line (recommended default)
• full - reads the entire response (code blocks are stripped)
• summary - a small headless Claude (haiku) condenses each response into a couple of spoken sentences and always includes any questions directed at the user. Best experience, but adds a few seconds of latency and a small per-turn API/usage cost. Mention both trade-offs before the user picks it.

5. Write the config

Write the choices to ~/.claude/auto-speak.conf in shell-variable form, e.g.:

VOICE="Tom (Enhanced)"
RATE=210
MODE="paragraph"

6. Test end to end

Run the plugin's hook script directly with a simulated payload so the user hears the real pipeline (use the actual plugin path from ${CLAUDE_PLUGIN_ROOT} if available, otherwise locate auto-speak.sh under ~/.claude/plugins/cache/):

echo '{"cwd":"'"$PWD"'","last_assistant_message":"Setup is complete. This is how your responses will sound from now on."}' | bash <path-to>/scripts/auto-speak.sh

Confirm the user heard it. Remind them: settings live in ~/.claude/auto-speak.conf (re-run this command anytime to change them), quick toggles exist (/auto-speak:mode, /auto-speak:voice, /auto-speak:rate, /auto-speak:timbre), and killall say stops playback mid-readout.

Finally, mention that Claude Code has a native /voice command (not part of this plugin) for talking to Claude instead of typing - paired with auto-speak, the conversation becomes fully hands-free in both directions.