Skip to content

[Improve] Keep voice work conversational during waits - #2574

Draft
roomote-roomote[bot] wants to merge 2 commits into
developfrom
feature/voice-progress-0b8tes5bu3f05
Draft

[Improve] Keep voice work conversational during waits#2574
roomote-roomote[bot] wants to merge 2 commits into
developfrom
feature/voice-progress-0b8tes5bu3f05

Conversation

@roomote-roomote

@roomote-roomote roomote-roomote Bot commented Sep 12, 2026

Copy link
Copy Markdown
Contributor

​Opened on behalf of @mrubens. View the task or mention @roomote-roomote for follow-up asks.

What changed

  • Have Fast deliver one short, activity-specific bridge before direct integration or delegated work likely to create noticeable silence, then let the existing voice path speak it.
  • Treat roughly 30 seconds without useful speech as a prolonged voice interval, while keeping updates concise, meaningful, deduplicated, and immediately interruptible.
  • Run the existing Session-scoped follow-through every 30 seconds during voice calls and every ten minutes otherwise, while unchanged checks remain silent.
  • Keep Fast's enforced work-start gate closed until that activity bridge has been delivered.

Why this change was made

Voice calls need to explain noticeable waits without narrating internal machinery or filling silence with repetitive status. This builds on the existing own-task follow-through path so direct work and delegated coding updates follow the same user-centered reporting semantics.

Impact

People hear what useful activity is beginning before the work starts, receive brief updates after prolonged silence or meaningful state changes, and can interrupt immediately. Text-mode acknowledgement and ten-minute follow-through behavior remain unchanged. The existing start and stop cues remain in place; no additional cue subsystem was added because spoken bridges already cover the state and the current cue path has no independent speech-aware ducking control.

@roomote-community

roomote-community Bot commented Sep 12, 2026

Copy link
Copy Markdown
Contributor

No new code issues found. See task

  • Voice-mode work can bypass the backend acknowledgement gate before the Live bridge is confirmed as spoken.
  • The retained one-minute voice wakeup cannot enforce the new 30-second silent-interval threshold.

Reviewed 7808103

Comment thread packages/cloud-agents/src/server/fast-agent/fast-agent-prompt.ts Outdated
Comment thread packages/cloud-agents/src/server/fast-agent/fast-agent-service.ts Outdated
@roomote-roomote

roomote-roomote Bot commented Sep 12, 2026

Copy link
Copy Markdown
Contributor Author

Fixed both review findings in 7808103a: Fast now delivers the activity bridge before its work-start gate opens, and voice follow-through now checks every 30 seconds so the silence threshold is enforceable. Focused voice, Fast, and wakeup tests plus type checks and pre-push validation pass.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant