dogfood Skill for Claude Code: Find Bugs in Your Mobile App

Makes Claude explore your iOS or Android app like a tester, find bugs and UX problems from real behavior, and capture evidence for each one.

Installs: 11.8K; on skills.sh
Stars: 5.2K; on GitHub
Updated: Nov 2026; last repo commit

dogfood is a short router for exploratory testing with the agent-device CLI. Claude opens your app, takes an interactive snapshot and a screenshot, explores flows, captures evidence for each issue, then closes the session. It triggers on phrases like dogfood, QA, exploratory test, bug hunt or test this app on mobile.

It needs the agent-device command (0.14.0 or newer) on your PATH. The SKILL.md has Claude read agent-device help dogfood for the current report format, and says findings must come from runtime behavior, not from reading source. It also bars Claude from running npm install -g agent-device@latest or npx -y agent-device@latest on its own. The folder is one SKILL.md and no scripts.

Good for

Mobile developers who want exploratory QA on a simulator, emulator or device
Teams who want bug reports backed by screenshots, not source reading
Claude Code users who already use the agent-device CLI

Install dogfood

Claude Code

1
Download the skill folder from https://github.com/callstack/agent-device/tree/HEAD/skills/dogfood.
2
Put it at ~/.claude/skills/dogfood/ so that SKILL.md sits directly inside.
3
Start a new session; the skill loads when its description matches the task.

Install the CLI yourself: the skill says Claude must not run npm install -g agent-device@latest or npx -y agent-device@latest on its own, and needs agent-device 0.14.0 or newer. The README says to run doctor yourself before handing the CLI to an agent. This skill is only the QA router; the main skill for driving apps is a separate skill, agent-device.

Built from the package data we hold. Not yet tested by us.

What's in the skill

When it loads

Systematically explore and test a mobile app on iOS/Android with agent-device to find bugs, UX issues, and other problems. Use when asked to dogfood, QA, exploratory test, find issues, bug hunt, or test this app on mobile.

Key facts

Author: Callstack
Repo: callstack/agent-device
License: MIT
Installs on skills.sh: 11,797 (Nov 2, 2026)
GitHub stars: 5,222 (Nov 2, 2026), whole repo
Last commit: Nov 1, 2026 (repo)
Files in the skill folder: 1 file including SKILL.md
Needs: The agent-device CLI, 0.14.0 or newer, and a target app on a simulator, emulator or device
Allowed tools: Bash(agent-device:*) and Bash(npx agent-device:*)
Works in: Claude Code, Codex, Cursor, Windsurf, Cline, Goose and any agent that can run a CLI

What it includes

Runs agent-device --version and stops if the CLI is missing.
If it fails, Claude tells you to expose a trusted binary or approve an exact-version npm command.
Runs code · Version check
Requires agent-device 0.14.0 or newer.
Older versions lack the help topics. Claude stops and asks you to upgrade.
Minimum version
Runs agent-device help dogfood for report shape and evidence commands.
The skill defers exact workflow details to the CLI's own help.
Runs code · Read the guidance
Open the app, snapshot and screenshot, explore flows, capture evidence, close.
A target app is required; Claude infers the platform or asks.
Runs code · Exploration loop
Findings come from runtime behavior, not source reads.
So each reported issue should be something Claude saw in the running app.
Evidence rule

From the official docs

Permissions

Medium risk

Claude can tap, type and confirm alerts in the app it tests, using the agent-device CLI. Use a simulator or a test account, not real customer data.

What it can do

Read your data
Create and change things
Delete things
Read-only mode

Watch out for

It acts in a live app
The README shows the CLI pressing buttons and filling fields, such as adding a contact. A bug hunt can create or change data in the app you test. github.com
It may call npx
allowed-tools includes npx agent-device, which can fetch the package. The SKILL.md bars autonomous npm install -g @latest or npx -y @latest and asks you to approve an exact version. github.com
Evidence can show private data
The SKILL.md has Claude capture evidence for each issue, such as screenshots. Check it before sharing, since a screen can show private data. github.com

Technical details

Scopes	Bash(agent-device:) Bash(npx agent-device:)

Questions

What does dogfood do?

It makes Claude test a mobile app by using it. Claude opens the app through agent-device, tries flows, and reports bugs and UX problems with evidence such as screenshots.

How do I install it?

Install the agent-device CLI yourself first, then add the skill. The install box shows both steps. The skill does nothing without the CLI.

Does it need an API key?

No key is mentioned. It needs Node.js 22.12 or newer for the CLI and a simulator, emulator or device with your app.

Which platforms does it test?

The skill says iOS and Android. The agent-device README lists more targets, such as tvOS, Android TV and HarmonyOS, with support depth varying by target.

Is it safe?

The skill bundles no scripts and limits Claude to agent-device and npx agent-device commands. But Claude taps, types and handles alerts in a real app, so use a test account and a simulator where you can.

dogfood vs the agent-device skill?

dogfood is a router for exploratory QA. The repo also has an agent-device skill for general app driving, which the dogfood SKILL.md does not require.

Information

updated Nov 2, 2026

Repository: callstack/agent-device
Folder: skills/dogfood

callstack/agent-device skills.sh

Sources

Docs, issue trackers and discussions this page is built from.

Show all 2

agent-device README · github.com
dogfood SKILL.md · github.com