Project 01
Agent Ally
/ Overview
AI coding agents pause mid-task to ask for permission: Can I run this command? Which option do you want? In the terminal those prompts use arrow-key widgets and streaming output that screen readers handle poorly, yet they are the highest-stakes moments in a session.
Agent Ally wraps the agent's decision loop in a fully accessible experience. Every decision is announced, navigated by keyboard, and confirmed aloud before anything commits. You can also approve decisions from your phone over Wi-Fi, so you are never tied to the terminal. The name is a pun: a11y (accessibility) + ally (a helper on your side).
/ Key Features
Spoken decisions: every permission prompt and multiple-choice question is read aloud with an earcon, full text, then options as a numbered list
Keyboard navigation: arrow and number keys move between options. Enter commits (high-risk needs a second Enter), R replays, D reads the full command
Phone control: scan a QR at launch, then send prompts and approve decisions from a web page over local Wi-Fi with no app to install
Phone voice: speak your prompts and have decisions read aloud from the phone
File and photo uploads, multiple chats, cross-device history, and session resume
/ Tech Stack
TypeScriptNode.jsWebSocketExpressWeb Speech APIClaude Code
/ My Role
Solo project. Designed and built the entire accessibility layer, the CLI wrapper around the agent, and the phone-control web app.