一键在 Manus 中运行任何 Skill

zeeme-mobile-voice-qualification

星标6

分支2

更新时间2026年3月7日 05:48

Qualify ZeeMe live voice reliability on iPhone, Samsung/Android, mobile browsers, and PWA builds before release. Use when mobile sessions show false barge-in interrupts, missed normal-volume speech, transcript gaps, or behavior that differs from desktop.

安装

用 Codex 或 Claude 帮你安装复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。

在 Manus 中运行

来源

CheickDiakite-yikes

CheickDiakite-yikes/my-ai-companion

打开 GitHub 仓库查看创作者相关仓库

下载

在 Manus 中运行

ZeeMe Mobile Voice Qualification

Run This Workflow

Capture at least 3 traces per platform with ?liveDebug=1:
- normal speaking volume, quiet room
- normal speaking volume, mild handling noise (phone movement/touch)
- intentional barge-in while assistant is speaking
Run:
- skills/zeeme-mobile-voice-qualification/scripts/mobile_voice_qualify.sh <trace1.json> <trace2.json> ...
Compare output with references/mobile-qa-matrix.md.
Tune at most 2 mobile sensitivity variables per iteration.
Re-run the same trace scenarios and require trend improvement, not one-off wins.

Qualification Gates (Mobile)

Pass only if all are true:
- user transcript chunks are present in each normal-speech scenario.
- activity_window_no_input_transcription <= activity_window_transcription_received.
- no repeated interrupted=true assistant cuts without user intent.
- interrupt button path remains healthy when explicitly used.
Fail and block rollout when:
- false interrupts dominate normal speech sessions.
- transcripts arrive only when shouting or exaggerated volume.
- Samsung/iPhone behavior diverges significantly without a known profile reason.

Tuning Order (Do Not Skip)

Mobile barge-in strictness:
- duration + peak/avg RMS minimums
- threshold frame requirement
Candidate speech stabilization:
- candidate clear timing and transition discipline
Only then evaluate broader VAD changes.

Keep Desktop Stable

Do not reuse aggressive mobile values on desktop.
Keep phone qualification separate from laptop acceptance.

Use Script Output

mobile_voice_qualify.sh provides:
- per-trace classification
- aggregate pass/fail counters
- likely failure mode (over-sensitive vs under-capture vs socket/interrupt drift)
- immediate next tuning direction

Use References

references/mobile-qa-matrix.md defines scenario matrix and release criteria.

同仓库更多 Skills

同仓库

zeeme-agent-gmail-context-setup

CheickDiakite-yikes/my-ai-companion

Configure and debug ZeeMe Gmail/Calendar personal context end-to-end with deterministic contracts for OAuth callback routing, token-encryption readiness, Zee Stage read/write flows, hands-free approvals, and voice/text smoke tests.

2026-03-136

zeeme-gemini-forensics

CheickDiakite-yikes/my-ai-companion

Diagnose ZeeMe Gemini integration failures using trace IDs, structured logs, and endpoint-specific checks. Use for live token errors, chat stream failures, live tool-response failures, Google-action approval drift, transcript issues, quota blocks, and model response anomalies.

2026-03-136

zeeme-incident-handoff

CheickDiakite-yikes/my-ai-companion

Build deterministic ZeeMe incident handoff packets from live-debug traces, branch context, and current status so another engineer can continue immediately. Use when voice/chat regressions are active, cross-environment drift exists, or ownership is changing between sessions.

2026-03-136

zeeme-live-voice-stability

CheickDiakite-yikes/my-ai-companion

Stabilize and debug ZeeMe live voice sessions that feel clipped, interrupted, delayed, or silent. Use when Gemini Live behavior is inconsistent across mobile/web environments and you need a trace-first diagnosis plus deterministic settings for VAD, interruption handling, thinking budget, token limits, and client audio capture.

2026-03-136

zeeme-replit-release-orchestration

CheickDiakite-yikes/my-ai-companion

Coordinate ZeeMe Replit-first hotfix and production rollouts with deterministic branch-sync checks, risk classification, VITE rebuild enforcement, and explicit rollback sequencing. Use when shipping live voice/chat fixes to Replit without breaking stable paths.

2026-03-136

zeeme-remotion-campaign-pipeline

CheickDiakite-yikes/my-ai-companion

Produce ZeeMe marketing videos with Remotion, including master renders, lightweight previews, and social cutdowns. Use when creating launch demos, campaign assets, or cinematic product showcases.

2026-03-116

name	zeeme-mobile-voice-qualification
description	Qualify ZeeMe live voice reliability on iPhone, Samsung/Android, mobile browsers, and PWA builds before release. Use when mobile sessions show false barge-in interrupts, missed normal-volume speech, transcript gaps, or behavior that differs from desktop.

ZeeMe Mobile Voice Qualification

Run This Workflow

Capture at least 3 traces per platform with ?liveDebug=1:
- normal speaking volume, quiet room
- normal speaking volume, mild handling noise (phone movement/touch)
- intentional barge-in while assistant is speaking
Run:
- skills/zeeme-mobile-voice-qualification/scripts/mobile_voice_qualify.sh <trace1.json> <trace2.json> ...
Compare output with references/mobile-qa-matrix.md.
Tune at most 2 mobile sensitivity variables per iteration.
Re-run the same trace scenarios and require trend improvement, not one-off wins.

Qualification Gates (Mobile)

Pass only if all are true:
- user transcript chunks are present in each normal-speech scenario.
- activity_window_no_input_transcription <= activity_window_transcription_received.
- no repeated interrupted=true assistant cuts without user intent.
- interrupt button path remains healthy when explicitly used.
Fail and block rollout when:
- false interrupts dominate normal speech sessions.
- transcripts arrive only when shouting or exaggerated volume.
- Samsung/iPhone behavior diverges significantly without a known profile reason.

Tuning Order (Do Not Skip)

Mobile barge-in strictness:
- duration + peak/avg RMS minimums
- threshold frame requirement
Candidate speech stabilization:
- candidate clear timing and transition discipline
Only then evaluate broader VAD changes.

Keep Desktop Stable

Do not reuse aggressive mobile values on desktop.
Keep phone qualification separate from laptop acceptance.

Use Script Output

mobile_voice_qualify.sh provides:
- per-trace classification
- aggregate pass/fail counters
- likely failure mode (over-sensitive vs under-capture vs socket/interrupt drift)
- immediate next tuning direction

Use References

references/mobile-qa-matrix.md defines scenario matrix and release criteria.