Skip to main content
prompt GP ChatGPT

channels — Claude Code prompt: what it does, how to install

Prompt for Claude Code and ChatGPT manages screen and channel selection between mobile apps and built-in browser.

Apache-2.0 ★ 9,488 updated 2026-11-16

Install into Claude Code · Manually: put the files in place All methods and platforms ↓
.claude/commands/channels.md
Open on GitHub

What channels does

This prompt instructs the agent on how to handle multiple screens, separating phone applications from built-in browser web pages. Each screen maintains its own tools and snapshot references. The agent follows a clear priority hierarchy: direct user requests first, app accounts for personalized data, and web browsing for fast lookups.

When a route hits a sign-in wall or blocks the browser, the system allows a single switch to the alternative channel. Confirmation and obstacle reports automatically target the most recently used surface or respect the specified surface argument.

What it helps with

  • choose between a web page and a mobile application
  • switch between phone tools and browser tools
  • handle browser blocks and sign-in walls on websites
  • open external app links on the phone

Example prompts for the agent

  • Choose the right channel for the user account task
  • Switch to the web browser and read the page data
  • Open the mobile app package to handle the external link

How to install channels

Manually: put the files in place

.claude/commands/channels.md

Questions about channels

How to install channels in Claude Code?

Place the prompt file into .claude/commands/channels.md within your project repository.

What does channels do?

It guides the AI agent on switching between mobile apps and web pages, managing screen-specific tools and references during tests.

How does the agent decide between app and web?

The agent prefers apps for account data, uses the web for fast searches, and switches once if a browser block or sign-in wall occurs.

Source text

channels.md · Apache-2.0
## Several screens: choosing a channel

This time you have more than one screen: apps on the phone, and web pages in the built-in browser. Each screen has its own set of tools (phone_ tools act on the phone, web_ tools act on web pages), and each keeps its own refs: a snapshotId from one screen cannot be given to the other screen's tools. To move to the other screen, just call its tools; there is no need to switch first. Confirmation, handover and obstacle reports refer by default to the screen you acted on most recently; when it is the other one, say so with the surface argument.

### Web page or app
At the start of the task you may get an "available channels" line saying whether the app is installed, whether there is a website, whether there is a known way to do it on the web, and whether the user has signed in to the site in Mika. Choose in this order:
1. When the user named a channel ("on the web", "in the app"), use only that one; if it does not work, say so plainly.
2. When the task needs things in the user's account (orders, messages, saved items, posts), prefer the app, because the user is usually signed in there. When the app is not installed, or the user has signed in to this site in Mika, use the web.
3. When the task only looks something up and there is a website, prefer the web: reading a whole page is fast and cheap, and there are no splash ads.
4. When the route you chose hits "this browser is not supported", a forced app download, or a sign-in wall while the user is away, switch to the other route, once only. When a site refuses the built-in browser, open the app if it is installed; otherwise end_not_done. This case does not need the user to take over.
5. Do not open a web page through another app, or launch an app from a web page, to get around an obstacle. A web result of external_app_link means the link wants to open an app: when the task really needs that app (not to get around an obstacle), open the package it gives on the phone instead of tapping the link again and again.

© wangzan101 · Apache-2.0 · wangzan101/Mika

Similar prompts

All prompts for Agent setup: memory, context, orchestration →

From the same set132

The whole set →
stop_summary stop_summary — a prompt for Claude Code and ChatGPT that summarizes progress when an assistant is stopped midway. It turns session logs into a concise status report in five lines or fewer. GP ChatGPT ★ 9,488repo stars web_script The web_script Claude Code prompt runs JavaScript to handle several tasks on the current web page. It can read list entries, move through results, or fill multiple fields in one call. GP ChatGPT ★ 9,488repo stars phone_sms_login phone_sms_login — a prompt for Claude Code and ChatGPT logs into an Android app via SMS on explicit user request. Device credentials stay local and never reach the model. GP ChatGPT ★ 9,488repo stars notices_read notices_read — a Claude Code prompt that reads notifications from apps Mika can access. Ask what came in or get a digest for a day, week, or month. GP ChatGPT ★ 9,488repo stars handover The handover prompt for Claude Code and ChatGPT tells an agent how to complete authorized tasks and hand control back to the user. It covers actions that need a password, one-time code, or CAPTCHA. GP ChatGPT ★ 9,488repo stars notice_day The notice_day prompt for Claude Code turns a list of app notifications into a structured daily digest. GP ChatGPT ★ 9,488repo stars notice_week The notice_week Claude Code prompt turns daily digests into a weekly account. It groups entries about the same matter and puts unresolved items first. GP ChatGPT ★ 9,488repo stars notice_month The notice_month prompt for Claude Code and ChatGPT turns weekly digests into a month-end recap. It combines notes about the same matter and tracks how each one changed. GP ChatGPT ★ 9,488repo stars

Project sponsors

Companies that keep this analytics open