Works with the AI Agents you already use
Gemini Live Voice Apps: Voice-First Editing That Never Wrecks Your Work
Build voice-first apps on the Gemini Live API where speech drives small, validated edits instead of full redraws, with the API key kept out of the browser and sessions that survive disconnects.
$12
Gemini Live Voice Apps: Voice-First Editing That Never Wrecks Your Work
Example session with this skill installed
Users of our voice whiteboard (Gemini Live in the browser) complain that every time they say "add a cache in front of the database" the whole diagram gets redrawn and their layout is lost, and once it deleted everything. Our tools and client setup are in tools.json and client.md. Redesign the tools in tools_v2.json and explain the fixes in FIX.md.
- Read your context and instructions
- Compiled the gemini live voice
tools_v2.json: edit_diagram with add/update/remove nodes and edges on stable ids plus confirm_large_removal; create_diagram only for an empty page or with new_page/replace; get_diagram for when the model is unsure; clear removed. FIX.md: every tool result returns {ok, errors, state} with ids, labels and connections; batches validated atomically and applied as one undo step; the API key moves to a server that mints single-use ephemeral tokens locked to the model, prompt and tools; the client stores session resumption handles and reconnects on goAway instead of asking users to reload.
Connects securely to your tools. The creator never sees your data.
About this skill
Voice changes the failure modes of an agent. Users speak in fragments ("put a cache in front of that"), correct themselves, interrupt, and judge the app by whether what they already built stays put. A voice editor that regenerates everything on every utterance feels broken even when each output is fine. This skill gives your agent the design that works.
What your agent builds with it
- Ephemeral-token server so the Gemini API key never reaches the browser: single-use tokens locked to your model, prompt and tools, plus endpoint hardening (
live_token.pybuilds the exact request and checks client setups against it). - One shared session setup with transcription, context-window compression and session resumption.
- Edit tools, not generate tools: typed operations on stable ids, every tool result returns the current state, new elements placed next to their neighbours.
- Atomic, validated edit batches (
edit_batch.py): whole-batch validation with actionable errors, removals before additions, spoken-name resolution ("the database", "this"), and a large-removal guard that stops "it deleted my diagram". - Realtime plumbing: audio formats, interruptions, tool responses,
goAwayand reconnects. - Stability evals: multi-turn scenarios against the real model that fail when untouched elements change.
Included: SKILL.md, Live protocol reference, live_token.py and edit_batch.py (standard library, no network), and an eval case with fixtures.
Part of the Production AI Agents series. Pair with Agent Tool Design principles and Agent Evals; start free with Agent Architecture.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for LLM Ops.
Live API message names, limits and model ids change; the reference is dated and tells you to verify. The edit engine is a reference implementation in Python to port to your client language, not a UI library.
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 1 day ago
- Accounts
- Gemini API key (server-side)
- Software
Python 3.10+
- Optional
Node.js
orPython web server for tokensBrowser with Web Audio
Permissions
File Scopes
Bundled scripts build JSON and validate edits locally; they never call the Gemini API.