[FEATURE] On-demand screen sharing in voice mode on iOS

Preflight Checklist

• [x] I have searched existing requests and this feature hasn't been requested yet
• [x] This is a single feature request (not multiple features)

Problem Statement

Voice mode on the Claude iOS app has no way to show Claude what's currently on the phone screen while staying in the conversation. To give Claude visual context, the user has to attach a photo manually, and both available paths are broken: the "+" button inside voice mode shows the image as attached on screen but Claude never receives it, and using the iOS share sheet from another app kills the active voice session and opens a brand new conversation instead of attaching to the one already running. As a result, there is currently no working way to combine voice mode with visual context on iOS.

Note: this is about the Claude mobile app (voice mode), not Claude Code CLI. Filing here for visibility since related mobile issues exist in this repo too.

Proposed Solution

Add on-demand screen sharing to voice mode on iOS, using a Broadcast Upload Extension (ReplayKit), started once by the user from the system picker or an in-app toggle. Claude should only capture a frame when explicitly triggered by voice (e.g. "look at my screen now"), not as a continuous stream. This keeps the feature lightweight on battery and privacy while giving Claude visual context it currently has no way to get during a live voice session.

Alongside this, fix the two attachment bugs blocking the same use case today: image sends from voice mode should either succeed and be confirmed, or fail visibly to the user instead of silently dropping; and share sheet input should attach to the active voice conversation instead of always starting a new one.

Alternative Solutions

Continuous video streaming during the whole voice session (like a video call) was considered but rejected in favor of on-demand capture: it uses significantly more battery and bandwidth, raises bigger privacy concerns since Claude would see everything continuously, and most use cases (asking about a specific screen or app) only need a single frame at the moment of the question, not a live feed.

Priority

Critical - Blocking my work

Feature Category

Interactive mode (TUI)

Use Case Example

User is in voice mode on their iPhone, walking through setting up a new app they've never used. They navigate to a settings screen they don't understand and say "look at my screen, what does this toggle do?" without touching the keyboard or leaving voice mode. Claude sees the current screen and answers out loud. The user keeps navigating and asking follow-up questions the same way, similar to asking a person looking over their shoulder.

Additional Context

ChatGPT's Advanced Voice Mode has supported on-demand screen sharing on iOS since December 2024 via the same ReplayKit APIs, so this is proven to work within Apple's platform restrictions — it's a product gap, not a technical barrier.

Related open issues in this repo point at the same underlying want (full parity with ChatGPT's live voice experience):

• #50720 — JARVIS-style voice-to-voice hands-free mode
• #57470 — voice conversation mode for Dispatch on iOS

This issue is specifically about the Claude mobile app's voice mode, not Claude Code CLI.