Skip to article frontmatterSkip to article content
Site not loading correctly?

This may be due to an incorrect BASE_URL configuration. See the MyST Documentation for reference.

Things on this page are fragmentary and immature notes/thoughts of the author. Please read with your own judgement!

AI agent CLIs frequently block waiting for the user’s input (approving a command, answering a clarifying question, etc.). Ideally, you get notified when an agent is waiting and reply directly from your phone without opening the terminal or even the session.

First-party Support

  1. Antigravity CLI/IDE sessions can be viewed and controlled at Antigravity Web/Hub .

  2. Claude Code has the most complete first-party support. See Remote Control vs cloud sessions for a comparison of all options.

    • Remote Control connects claude.ai/code or the Claude mobile app (iOS/Android) to a Claude Code session running on your machine. Code execution and filesystem access stay local. Start it with claude remote-control (server mode), claude --remote-control (--rc), or /remote-control (/rc) inside a running session. Press spacebar (in server mode) to show a QR code for opening the session in the Claude app. Run /config and set Enable Remote Control for all sessions to true to turn it on automatically. Available on Pro, Max, Team and Enterprise plans (not with API keys). On Team/Enterprise, an admin must enable it first.

    • Mobile push notifications: when Remote Control is active, run /config and enable Push when actions required (permission prompts and questions) and/or Push when Claude decides (e.g., a long-running task finishes). You can then approve or reply from the Claude app.

    • Channels (research preview) push events from chat apps (Telegram, Discord, iMessage) into a running Claude Code session, and Claude can reply back through the same chat. Channels supporting permission relay forward permission prompts to the chat so that you can approve or deny there. This is exactly “reply to a chat message to unblock the agent”. E.g., claude --channels plugin:telegram@claude-plugins-official.

    • Dispatch lets you message a task from the Claude mobile app to the Claude Desktop app on your machine.

    • Claude Code on the web (cloud sessions) runs tasks on cloud infrastructure, which can be started and monitored from the browser or the Claude mobile app.

    • Claude Code in Slack: mention @Claude in a channel to start a cloud session.

  3. OpenAI Codex is available in the ChatGPT mobile app (iOS/Android). Pair your phone with the Codex app on your computer by scanning a QR code, and then you can monitor threads, approve commands and start new tasks from the phone. The host machine has to stay awake and online.

Third-party Tools

  1. Happy is an open-source, end-to-end encrypted mobile/web client for Claude Code, Codex and other agent CLIs (Antigravity, any ACP CLI, etc.). Run happy (or happy claude) in place of claude, or happy codex for Codex, and scan the QR code to connect from your phone or browser.

  2. Clawless bridges local agent CLIs (Gemini CLI, OpenCode, Claude Code) to Telegram or Slack via ACP (Agent Client Protocol). You chat with the agent directly in Telegram/Slack.

  3. HumanLayer used to provide approval workflows (e.g., an MCP server for Claude Code) which route approval requests to Slack, email, etc. The repo is deprecated now and the product has pivoted to a multiplayer coding agent IDE/cloud (humanlayer.com).

DIY Approaches

  1. ntfy supports action buttons in push notifications. An http action button can call a webhook you run which then feeds the response to the agent. Anyone knowing the topic can trigger the webhook, so use a hard-to-guess topic and authenticate the webhook.

    curl \
    -H "Actions: http, Approve, https://your-webhook.example/approve; http, Deny, https://your-webhook.example/deny" \
    -d "Agent is requesting permission to run: rm -rf ./cache" \
    ntfy.sh/your-secret-topic

  2. Use hooks of AI agent CLIs (e.g., the Notification hook of Claude Code) to send a message to Telegram/Slack/ntfy when the agent is waiting for input. Note that this is one-way only. You need something else (e.g., a bot listening for replies) to send responses back.

  3. Run agents in a terminal multiplexer (tmux, Zellij) and have a background bot watch pane output for prompts (e.g., (y/n)). The bot notifies you and types your reply into the pane, e.g., via tmux send-keys -t agent:0 "your response" Enter. This works with any CLI tool but is fragile.

References

Oct
Jujutsu Equivalent of Git Stage
Oct
Tips on the Moli Browser for Agents