- Home
- Skills
- Prompt & Skill Engineering
- Voice Replies
More screenshots
Works with the AI Agents you already use
Voice Replies
Format agent replies for natural-sounding text-to-speech by separating spoken summaries from on-screen code and data.
$5
Voice Replies
Example session with this skill installed
Fix the bug in src/utils/auth.ts where the token expires too early and let me know when it's done.
- Read your context and instructions
- Compiled the voice replies
SAY: I updated the auth utility to keep you logged in for twenty-four hours.
I modified src/utils/auth.ts to update the EXPIRATION_BUFFER.
// src/utils/auth.ts
- export const TOKEN_EXPIRY = 3600; // 1 hour
+ export const TOKEN_EXPIRY = 86400; // 24 hours
Next steps
- Refresh your browser to pick up the new token duration.
- Check the server logs if the session still drops unexpectedly.
Connects securely to your tools. The creator never sees your data.
About this skill
Standard LLM answers are written for the eye, not the ear. When a text-to-speech engine meets Markdown, file paths or abbreviations, it reads them out character by character and the answer sounds broken. This skill teaches your agent to keep what it says apart from what it shows, so your voice assistant sounds natural.
What it does
- Splits spoken and on-screen output. One line starting with
SAY:is spoken; code, paths and details go after it, on screen. - Writes for the ear. The spoken line has at most two short sentences, no Markdown, paths or symbols, small numbers as words ("three files"), and "for example" instead of "e.g.".
- Stays quiet when nothing needs saying. After a simple action that worked, such as opening an app or playing a song, the
SAY:line is left empty. - Expects speech-to-text mistakes. The agent treats requests as transcripts and picks the sensible meaning ("get push" means "git push").
- Asks one clear question. An optional
ASK:line carries a short yes/no or either/or question, so your app can read it out and open the mic. - Matches the user's language. Hindi and Hinglish replies are written in Roman letters, never Devanagari.
How it works
- The agent reads the user's request as a speech transcript and works out what was meant.
- It does the work with its tools. It never puts a
SAY:line in a message that a tool call follows, so only the final answer is spoken. - Its final message starts with one
SAY:line, plus anASK:line when it needs an answer. - Code, logs and file names follow on screen for the user to read.
Frameworks & tools
Written for voice assistants built on Claude Code or Codex. It works with any text-to-speech engine, as long as your app reads the SAY: and ASK: lines. Includes tips for speaking while the reply is still streaming and for replies that come back with no SAY: line. Drawn from Nova, an open-source voice assistant.
Why this beats prompting it yourself
You could write these rules yourself, but you'd find them one awkward spoken reply at a time. This skill has them already worked out and used in a real voice assistant: the reply format, the "silence means it worked" rule, the list of what never to read aloud, and bad-versus-good examples.
Use cases
- Hands-free coding assistants for developers.
- Voice control for desktop, smart home or server tools.
- Voice-first workflows where the screen is only for reference.
Known limitations
The agent follows these rules most of the time, not always, so your app should handle a reply with no SAY: line (speak only its first sentence). Your app must read the SAY: and ASK: lines and run the text-to-speech itself. Details on screen are never spoken, which does not suit users who can't see the screen. Hindi replies come in Roman letters, so your voice must pronounce romanized Hindi well.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Prompt & Skill Engineering.
Requires a host application capable of parsing the SAY: and ASK: prefixes to separate audio from UI. Does not support Devanagari; Hindi must be written in Roman letters.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 5 days ago
Nothing to install. It runs in your AI agent.
Your voice app must speak the SAY: line and open the mic for the ASK: line. The skill only shapes the agent’s replies.
Permissions
No special permissions declared or detected