Voice Agent Skills

Thirteen provider-neutral skills that help a coding agent build, debug and review AI voice agents people enjoy talking to. They cover the problems that decide whether a call feels natural: when the agent should speak, how fast it answers, what happens to the audio, and whether the call stays legal and secure. Each skill works with any stack, from managed platforms like Vapi or Retell to frameworks like LiveKit Agents or Pipecat.

Skill	Use it when
voice-stack-selection	Choosing a platform, speech engines, transport and hosting
voice-turn-taking	Callers get cut off, the agent talks over them, or replies come late
voice-conversation-design	Writing prompts for spoken turns, clarification and handoffs
voice-speech-pipeline	Picking and tuning speech-to-text and text-to-speech
voice-audio-frontends	Echo, noise, gain or competing speakers
voice-latency-audit	The agent feels slow and you need to find where the time goes
voice-media-debugging	Silent, one-way, distorted or delayed audio
voice-call-reliability	Dropped calls, failed transfers or duplicate actions
voice-data-capture	Getting phone numbers, emails, dates and codes right
voice-agent-evaluation	Testing a voice agent before and after launch
voice-cost-estimation	Working out cost per minute, per call and per month
voice-phone-compliance	Recording consent, outbound calling rules and number reputation
voice-agent-security	Caller verification, prompt injection and sensitive data

Ask for what you need in plain words, such as "why does my agent keep interrupting people?", and the matching skill loads on its own.

The skills give engineering guidance, not legal advice. Check compliance questions with a qualified professional for your jurisdiction.

Guides, diagrams, a landscape of voice AI products and an index of official provider skills live in the main repository: github.com/RobStrayer/voice-agent-skills.

MIT licensed. Report problems or suggest changes in the issue tracker.