voice

Voice Integration

Maps Claude Code's native /voice mode to Maestro operations. When voice mode is active, spoken commands are transcribed and routed to the appropriate Maestro command.

Voice Command Patterns

When the user speaks, Claude Code transcribes the audio. This skill helps Claude interpret voice input as Maestro commands:

Voice Input (examples)	Maps To
"Build [feature]" / "Create [feature]"	/maestro "[feature]"
"What's the status?" / "How's it going?"	/maestro status
"Show the board" / "Show my stories"	/maestro board
"Plan [feature]" / "Let's plan [feature]"	/maestro plan "[feature]"
"Initialize" / "Set up Maestro"	/maestro init
"Help" / "What can you do?"	/maestro help
"Run diagnostics" / "Health check"	/maestro doctor
"Show settings" / "Show config"	/maestro config
"Change models" / "Update model"	/maestro model
"Show history" / "Past sessions"	/maestro history
"Search [topic]" / "Find in brain"	/maestro brain search "[topic]"
"Save a note" / "Remember this"	/maestro brain save
"Daily briefing" / "What's happening?"	/maestro brain daily
"Pause" / "Stop for now"	/maestro status pause
"Resume" / "Continue"	/maestro status resume
"Abort" / "Cancel"	/maestro status abort
"Send notification" / "Notify team"	/maestro notify "[message]"
"Show diagram" / "Visualize"	/maestro viz
"Start opus" / "Build product"	/maestro opus "[vision]"

TTS-Optimized Responses

When voice mode is active, Maestro should optimize its output for spoken delivery:

Rules for voice output:

1. Keep it short: Max 3 sentences per response
2. No tables: Convert to spoken lists ("First, ... Second, ... Third, ...")
3. No code blocks: Describe what was done, not the code
4. No box-drawing: Plain conversational text
5. Numbers spoken naturally: "about three dollars" not "$3.12"
6. Status as sentences: "Story three of five is done. QA passed on first try." not a table

Example voice responses:

Status query:

"You're working on user authentication. Three of five stories are done. The current story is the frontend, and it's being implemented right now. Total cost so far is about two dollars."

Story checkpoint:

"Story three is done. The frontend components are built and QA approved it on the first try. Want me to continue to the next story?"

Feature complete:

"Feature complete. All five stories passed. Total cost was four dollars and twenty cents. Would you like me to create a pull request?"

How to Activate

1. Start Claude Code normally
2. Press /voice to enter voice mode (push-to-talk)
3. Speak your command — Claude transcribes and Maestro routes it
4. Maestro responds in voice-optimized format

Detection

When a message arrives as [Voice transcribed]: prefix (from Claude Code's voice mode), or when the conversation context indicates voice mode is active:

1. Parse the transcribed text for Maestro command patterns
2. Route to the appropriate command/skill
3. Format the response for TTS (short, conversational, no tables/boxes)
4. If the command requires interactive choices (AskUserQuestion), present options verbally: "Would you like to continue, review the changes, or abort?"

Integration

With Dev-Loop

At checkpoint phase, if voice mode is active:

• Use TTS-optimized response (short, no tables)
• Read the story summary aloud
• Ask "continue, review, or abort?" verbally

With Status Command

When voice is active and user asks for status:

• Speak: "You are working on [feature]. [N] of [M] stories done. Current phase is [phase]."

With Notifications

Voice can complement notifications:

• Audio chime for checkpoint (audio skill)
• Spoken summary for feature completion (voice skill)

Configuration

voice:
  enabled: auto  # auto-detect when /voice is active
  tts_optimization: true
  max_response_sentences: 3