Consult Mode

Speak a checkpoint message — the user may want to respond, but no mic recording happens. Use for preemptive checkpoints before important actions.

When to Use

• Before destructive actions: "About to drop the staging database. Okay?"
• Before commits/pushes: "I'm ready to push to main. Want to review first?"
• Decision points: "Should I use Redis or the built-in cache?"

MCP Tool

Name: qa_voice_consult

Parameters

Parameter	Type	Required	Default	Description
message	string	Yes	—	The checkpoint question or status (non-empty after trimming)
rate	string	No	+5%	Speech rate override

Returns

{
  "content": [{
    "type": "text",
    "text": "[consult] Spoke: \"your message\"\nUser may want to respond. Use qa_voice_converse to collect voice input if needed."
  }]
}

Errors

Same as Announce — edge-tts or audio player issues.

Behavior

1. Text is spoken aloud via edge-tts (same as announce/brief)
2. Returns immediately — no mic recording
3. The return text hints that the agent should follow up with converse if input is needed

Consult is a "heads up" — it doesn't block. If the user says nothing (types nothing), the agent can proceed. If the user wants to respond verbally, the agent should call qa_voice_converse as a follow-up.

Speech Rate

Default: +5% (slightly faster than natural — checkpoints should be brisk).

When to Follow Up with Converse

The consult tool's response includes a hint:

"User may want to respond. Use qa_voice_converse to collect voice input if needed."

The agent should decide based on context whether to:

1. Wait for text input — if the user typically types
2. Call converse — if in a voice session and input is expected
3. Proceed — if the checkpoint was informational only