speak

Speak (Crux TTS)

This is the only path agents should use to produce voice. It always calls the Crux server (or tries to).

Command

bash scripts/speak.sh "Plain English only, no symbols" [priority] [persona]

• priority: 0 = queue with other speech (default). 1 = preempt current speech.
• persona: optional — forge, sentinel, chronicle, sage, maestro, echo, priya, siri (see CONTRIBUTING.md).

Wait for playback to finish before ending the turn:

bash scripts/speak.sh --sync "Your closing summary here" 1

Rules

• 050-no-markdown-in-speech — spoken strings must be plain English.
• 000-agent-http.mdc — voice table (start, major change, end, blocked).
• 010-acknowledge-before-planning.mdc — ack before work; batch speak.sh with first tool calls.
• 015-agent-speak-mandatory.mdc — mandatory spoken ack and completion; never skip speak.sh because /health failed.
• Use curl for non-speak APIs per CONTRIBUTING.md; do not hand-roll curl to /speak for routine narration.
• Do not call say, afplay, or local TTS.

Health

You are not responsible for server uptime. Always run speak.sh when the rules require it. If curl fails, note it in chat if relevant; do not use that as a reason to skip voice attempts.

Env

CRUX_BASE_URL if the server is not on http://127.0.0.1:9090.