Agent skill

Sim Use

by lycorp-jp in lycorp-jp/sim-use

Drive iOS Simulator, Android emulator/device, and physical iPhone/iPad screens for AI agents.

Apache-2.0Auto-check passedMobile

Install Sim Use

skills CLI
$ npx skills add lycorp-jp/sim-use --skill sim-use -a claude-code

Project install by default; add -g for ~/.claude/skills/.

GitHub CLI
$ gh skill install lycorp-jp/sim-use sim-use --agent claude-code

Project scope by default; add --scope user for a personal install. Needs GitHub CLI 2.90.0 or later (public preview).

Manual copy
$ git clone --depth 1 https://github.com/lycorp-jp/sim-use.git skills-src && mkdir -p .claude/skills && cp -r skills-src/skills/sim-use .claude/skills/sim-use && rm -rf skills-src

Use ~/.claude/skills/ instead of .claude/skills for a personal install. The folder must contain SKILL.md.

Claude Code skills documentation · loads skills from .claude/skills/

Facts

Skill name
sim-use
GitHub stars
1.4k
Token cost
~3.2k tokens
SKILL.md length
1,589 words
Files
6 (incl. scripts, references)
Skills in repo
4
Repo updated
First seen
Licence
Apache-2.0

At a glance

Drive iOS Simulator, Android emulator/device, and physical iPhone/iPad screens for AI agents.

  • Works in 6 steps: Preflight → The observe-act loop → Pitfalls → …
  • Asked to automate a simulator
  • SKILL.md covers 0. Preflight, 1. The observe-act loop, 2. Pitfalls and 3. Crash awareness, plus 2 more sections
  • Runs Python scripts from its folder; calls python3

What it does

Sim Use is an agent skill from lycorp-jp/sim-use. Drive iOS Simulator, Android emulator/device, and physical iPhone/iPad screens for AI agents. Use when asked to automate a simulator or emulator, drive a real iOS device, tap/swipe/type on a device, describe UI, take a screenshot, or interact with a mobile app.

Its SKILL.md is about 3.2k tokens, which your agent loads only when the skill is triggered. The skill folder holds 7 other files, including scripts and reference files (for example `references/batch-reference.md`, `references/cheatsheet.md` and `references/crash-awareness.md`).

It sits in Mobile. It works with iOS and Android. The repository describes itself as: Give your AI agent eyes and hands on iOS Simulator and Android emulator/devices. The licence is Apache-2.0.

When your agent uses it

  • Asked to automate a simulator
  • Drive a real iOS device
  • Tap/swipe/type on a device
  • Take a screenshot

Example prompts

  • “/sim-use”

Requirements

  • Python 3

Workflow steps

6 steps, taken from the step headings in SKILL.md.

  1. Preflight
  2. The observe-act loop
  3. Pitfalls
  4. Crash awareness
  5. Escalation
  6. Exit checklist

What it can do on your machine

Read from SKILL.md and the folder at commit 3cd5070. It shows what the files ask for, not the result of running them.

  • Tool permissions

    Pre-approves nothing: there is no allowed-tools line, so your agent's usual permission prompts apply.

    From allowed-tools in the SKILL.md frontmatter.

  • Runs code

    Ships 1 file in scripts/ (Python), which the agent can run.

    Shell commands in SKILL.md call:

    • python3

    From the folder's file list and the shell code blocks in SKILL.md.

  • Network

    No URLs in SKILL.md.

    From URLs in SKILL.md, links to its own repository left out.

  • Credentials

    Names no API keys, tokens, secrets or passwords.

    From names ending in _API_KEY, _TOKEN, _SECRET, _KEY or _PASSWORD in SKILL.md.

Context cost

Sim Use loads about 3.2k tokens when it runs, and up to ~8.7k if it reads all its reference files. Until then it costs about 67 tokens; SKILL.md has 1,589 words of instructions outside code blocks.

Always · name and description, kept in context so the agent knows when to use it
~67
When it runs · the whole SKILL.md, loaded when a task matches
~3.2k
With references · SKILL.md plus every file in references/, read only if the agent opens them
~8.7k

Estimates: characters ÷ 4, the usual rule of thumb; real counts depend on the model's tokenizer. Scripts and assets cost tokens only if the agent reads them.

Safety

Auto-check passed

The automated check found no risky patterns in SKILL.md.

Automated static check — not a guarantee. Review scripts before installing. It scans the text of SKILL.md for risky patterns (piping downloads into a shell, reading credential files, hidden Unicode, destructive commands); the scripts in this folder are not scanned.

SKILL.md

The full file from lycorp-jp/sim-use at commit 3cd5070, republished under its Apache-2.0 licence (© lycorp-jp). 1,589 words, ~3,172 tokens.

Download SKILL.mdSave it as .claude/skills/sim-use/SKILL.md (or your agent's skills folder). This skill also uses 5 other files; get the full folder from GitHub.
name
sim-use
description
Drive iOS Simulator, Android emulator/device, and physical iPhone/iPad screens for AI agents. Use when asked to automate a simulator or emulator, drive a real iOS device, tap/swipe/type on a device, describe UI, take a screenshot, or interact with a mobile app.

0. Preflight

Before first interaction with a device, run the preflight check:

bash
python3 scripts/preflight.py --device <UDID>

This verifies sim-use is installed and compatible with the skill, the device is reachable, and the daemon is healthy. If you don't have the script, do the checks manually:

  1. sim-use --version — confirm sim-use is on PATH and supports the skill's documented commands.
  2. sim-use devices — confirm the target device is listed and booted/connected.
  3. sim-use ui --device <UDID> — confirm you can read the screen.

A WARN UI content line still passes preflight: the read works, but the outline can miss app controls. See the Pitfalls index.

--device is optional when only one simulator is booted or one daemon is running. For Android, run sim-use android init --device <serial> once to install the bridge APK. Attached physical iPhones/iPads appear in sim-use devices with kind physical and route through the top-level verbs too — but only ui, selector-based tap and screenshot; every other verb rejects on that target. See Physical iOS devices below before driving one.

1. The observe-act loop

Every interaction follows the same cycle: observe → act → verify.

Observe
bash
sim-use ui --device <UDID>

Read the outline. Each element has an @N alias and optionally a #<id> identifier. List cells carry #N (dominant list) or #N@M (scoped).

Frames in the JSON output (--json: entries[].frame, screen) are in platform-native units — iOS points, Android pixels. Key off the envelope's platform field before doing math on coordinates across platforms. Always pair --json with --no-raw — see Keeping output small below.

Act

Pick a selector, in order of preference:

SelectorWhen to use
tap @NRight after ui. Fastest, cache-backed.
tap #<id>Stable across minor layout changes. Paste from the outline.
tap --label 'X'Scripted flows. Combine with --wait-timeout for transitions.
tap --label-regex '...'Dynamic labels with counters/timestamps. Anchor with ^...$.
tap --label-contains 'X'Substring match when exact label is unknown.
tap -x N -y N / tap --point x,yLast resort for elements with no AX data. Device-native portrait space by default; add --coordinate-space ui when the numbers come from the outline and the device may be rotated.

Disambiguate collisions with --element-type or --frame minY=0.7r (see references/cheatsheet.md).

Verify

Always verify after acting — commands are fire-and-forget:

bash
sim-use ui --device <UDID>       # read the new screen state
sim-use screenshot --device <UDID> --output after.png
Keeping output small

Every byte of command output you read costs context. Defaults that keep the loop cheap:

  • Prefer the default text outline over --json. The outline carries everything a tap needs (@N / #<id> aliases, roles, frames, states); reach for --json when you need structured fields for coordinate math (entries[].frame, screen) or full untruncated text (the outline truncates labels at 60 graphemes, value= at 30).
  • When you do use --json, add --no-raw. data.raw is the raw accessibility tree — typically the bulk of the envelope's bytes, and useful only for debugging sim-use itself.
  • One ui per action: the Verify read of step N is the Observe read of step N+1. Don't run a second ui in between.
  • Verify with the text outline, not a screenshot. Reading a screenshot costs several times more than a typical outline; take one only when the check is genuinely visual (colors, images, layout).
  • On iOS, to wait out a transition, prefer tap --label 'X' --wait-timeout 3 (polls for the element) over re-running ui in a loop. Android tap has no --wait-timeout; use sleep between commands instead.
  • For a known multi-step sequence on iOS, use sim-use ios batch (see references/batch-reference.md) — one invocation, one output.
Common moves
TaskCommand
Scroll downsim-use gesture scroll-up --device <UDID> (scroll-up = content moves up = page down)
Type textsim-use type 'hello' --device <UDID>
Paste unicodesim-use paste 'こんにちは 🎉' --device <UDID> (iOS: needs hardware keyboard)
Hardware buttonsim-use button home --device <UDID>
Android backsim-use button back --device <UDID>
Wait for animationsleep 0.4 between commands, or --pre-delay 0.5
Toggle/switchsim-use tap @N --duration 0.05 --device <UDID> (UISwitch needs a brief hold)
Swipesim-use swipe --from 50,500 --to 350,500 --device <UDID>
Pinch zoom insim-use gesture pinch-out --device <UDID> (two-finger spread)
Rotatesim-use gesture rotate-cw --angle 90 --device <UDID>
Record evidence GIFsim-use record-video --output demo.gif --device <UDID> — stop with SIGINT/SIGTERM (never SIGKILL); transcodes after stop; auto-plays inline in PRs; add --gif-markers for START/END loop-boundary cards
Physical iOS devices (experimental)

The top-level verbs route a physical UDID automatically, but only three of them: ui, tap (#<id> / --id / --label / --label-contains / --element-type forms) and screenshot. Never assume capability parity with the simulator — every other verb or form (coordinates, @N/#N aliases, swipe/gesture/multi-touch, type/paste, recording, --value/--label-regex/--frame/--duration/--wait-timeout) rejects with the reason and the nearest alternative in the hint; read it instead of retrying. The sim-use ios-device namespace is the physical-only peer of ios/android (same verbs, plus ECID addressing and tree-tuning flags).

Hard requirement for ui / tap: the device must be paired, trusted, unlocked and in Developer Mode, and the foreground target app must be development-signed with get-task-allow=true. A Release-configuration binary installed with a Development profile is supported. Distribution/Ad Hoc, TestFlight, App Store and system apps are unsupported; do not retry them or claim success. sim-use itself installs and signs no runner and needs no Developer Disk Image. screenshot is exempt from the signing rule — it runs over CoreDevice and captures any screen, system apps included.

bash
# Physical-device preflight — physical rows carry kind `physical`
sim-use devices

# Observe → act → verify, same loop and verbs as the simulator
sim-use ui --device <UDID>
sim-use tap --label "Friends" --element-type Button --device <UDID>
sim-use ui --device <UDID>

# For dynamic labels
sim-use tap --label-contains "Reply" --element-type Button --device <UDID>

# By stable identifier (the #id shown in ui) — positional or --id
sim-use tap '#BackButton' --device <UDID>

# Screenshot — any screen, not limited to development-signed apps
sim-use screenshot --output shot.png --device <UDID>

Rules for this experimental surface:

  1. Treat hierarchy errors as capability failures. If the command says the hierarchy is unavailable, confirm the screen is unlocked and inspect the installed app's final get-task-allow entitlement. Do not fall back to coordinates or focus walking.
  2. Tap by #id or label, not @N. Element handles expire with the DTX connection, so there is no @N alias (nor coordinates — no geometry). Use the #id shown in the outline (positional #<id> or --id) — it is stable and the best choice when a label is dynamic — or --label / --label-contains, with --element-type to disambiguate. The navigation-bar back button appears as a normal Button "<previous screen title>" #BackButton; go back by tapping #BackButton (or the shown label) like any other element — no special "back" verb.
  3. Always verify. Activate is fire-and-forget. Re-run sim-use ui and confirm the expected state before continuing.
  4. Respect the capability rejections. A not supported on physical iOS devices error is a statement about the channel, not a transient failure — follow its hint (usually ui + tap '#<id>' / --label) instead of retrying or substituting a lookalike form. --json works on every verb with the standard {ok, data} envelope; physical results carry "kind":"physical" and omit geometry fields (screen, x/y).
  5. Budget seconds, not milliseconds. A full tree takes a few seconds. sim-use ios-device ui --fast is quicker but omits nested elements; do not poll in a tight loop.

If ui succeeds with zero elements or tap prints success for a missing/ambiguous label, treat it as a sim-use bug; the command is expected to fail loudly instead.

Show full SKILL.md (488 more words)Show less

2. Pitfalls

Quick symptom index — see references/pitfalls.md for detailed recipes.

SymptomCauseFix
tap --label hits wrong elementLabel collision (e.g. header and tab bar share text)Add --frame minY=0.7r or --element-type to narrow
tap @N fails after navigationAlias cache is staleRe-run ui before tapping
App: line shows wrong appSystem layer (alert, share sheet) is on topDismiss it first, then re-run ui
multipleMatches errorSeveral elements share the selectorUse --frame, --element-type, or a more specific selector
Tap lands but nothing happensAnimation in progress, or element not yet interactiveAdd --pre-delay 0.3 or --wait-timeout 3
iOS: paste drops textSoft keyboard only; HID Cmd+V is ignoredUse paste --via-menu --target-id <id>
Android: paste deniedBackground clipboard access blockedUse type instead
Outline shows U+FFFC in labeliOS icon placeholder characterMatch with --label-regex excluding the prefix
[i] … covers ~N% of the screen warning (text output, or --json top-level advisory key)The selector resolved to a near-full-screen wrapper (common on Flutter/canvas UIs) and the tap hit its center, likely missing the intended controlRe-run ui and target the control via @N/#<id>, or pass explicit -x/-y/--point
[i] Screen orientation could not be confirmed… / …coordinates may be stale… advisoryDevice/app is rotated (the App: header shows a tag like (landscape-right)) and orientation self-calibration couldn't verify the mapping, or the @N snapshot predates a rotationRe-run ui and tap again; selectors handle rotation automatically once calibration succeeds. Explicit -x/-y/--point is device-native portrait space by default — on tap/swipe/touch, pass --coordinate-space ui to use outline (visual-space) coordinates on a rotated device
Outline is empty or misses visible app controls, or [i] … recovered from other processes … advisory (--json: advisory.kind: remote_content_recovery)The app exposed an empty accessibility tree. Normal for a system picker. Otherwise, on an iOS simulator, a likely cause is that app accessibility was off when the app launchedCompare the outline with the screen. If controls are missing, see Missing app controls

3. Crash awareness

See references/crash-awareness.md for the full protocol. Summary:

sim-use watches for the target process disappearing between commands. When it detects a crash:

================ PROCESS DISAPPEARED ================
com.example.app (pid 12345) was alive at the previous command and is GONE now.

On Android, ui also detects the AOSP system crash dialog directly from the accessibility tree.

Mandatory response:

  1. STOP. Do not silently relaunch or continue.
  2. Report the crash to the user with the banner text.
  3. Wait for instructions before proceeding.

After an intentional relaunch, call sim-use app-state --reset to clear the signal.

4. Escalation

Stop and ask the user when:

  • A selector collision cannot be resolved with available disambiguators.
  • Preflight fails and autofix does not recover.
  • The task requires a destructive action (deleting data, uninstalling an app).
  • You've retried the same action 3 times without progress.

5. Exit checklist

Before reporting a task as complete:

  1. Run sim-use ui (or screenshot) to capture the final state.
  2. Confirm the screen matches the intended outcome.
  3. If the outcome is ambiguous, show the final ui output or screenshot to the user.

© lycorp-jp, Apache-2.0. Rendered from Markdown: HTML in the file is shown as text, images as links, and headings moved down two levels. Raw file

Files

SKILL.md and 5 other files (scripts, references) in skills/sim-use of lycorp-jp/sim-use.

  • SKILL.md
  • references/batch-reference.md
  • references/cheatsheet.md
  • references/crash-awareness.md
  • references/pitfalls.md
  • scripts/preflight.py

Open the folder on GitHubat commit 3cd5070

Compare with similar skills

Sim Use next to the 5 skills that share the most tags, products or categories with it. Stars are the repository's; “used in” counts other GitHub owners with a copy.

Sim Use compared with similar skills
SkillStarsUsed inTokensAuto-checkLicenceRepo updated
Sim Use this skilllycorp-jp/sim-use1.4k—~3.2kAutomated safety check: PassApache-2.0
Mobilerun Docs Referencedroidrun/mobilerun9.6k—~943Automated safety check: PassMIT
App Store Screenshots GeneratorParthJadhav/app-store-screenshots7.2k—~14kAutomated safety check: PassMIT
Rnn Codebasewix/react-native-navigation13k—~2kAutomated safety check: PassMIT
Building Native UICherryHQ/cherry-studio-app4k8 repos~2.6kAutomated safety check: PassMIT
Appiumblokadaorg/blokada3.3k—~3.5kAutomated safety check: PassMPL-2.0

Similar skills

  • Mobilerun Docs Reference

    droidrun/mobilerun

    Answers questions about Mobilerun, the LLM-agent framework for automating Android and iOS devices, by pointing the agent to the right page of its v5 documentation.

    9.6k GitHub stars~943 tokensUpdated 2 days ago
    MobileAuto-check passed
  • App Store Screenshots Generator

    ParthJadhav/app-store-screenshots

    Scaffolds a Next.js editor for designing App Store and Google Play screenshots as ads and exporting them at every required size, for iOS, Mac and Android.

    7.2k GitHub stars~14k tokensUpdated 3 days ago
    MobileAuto-check passed
  • Rnn Codebase

    wix/react-native-navigation

    Official

    Navigate and work with the react-native-navigation (RNN) codebase.

    13k GitHub stars~2k tokensUpdated yesterday
    MobileAuto-check passed
  • Building Native UI

    CherryHQ/cherry-studio-app

    Complete guide for building beautiful apps with Expo Router.

    4k GitHub starsUsed in 8 repos~2.6k tokens
    MobileAuto-check passed
  • Appium

    blokadaorg/blokada

    A skill your agent uses for dynamic inspection and navigation of the Blokada app through the repo-local Appium machine session.

    3.3k GitHub stars~3.5k tokensUpdated today
    MobileAuto-check passed
  • Phoneagent

    rounak/PhoneAgent

    Control a connected iPhone, iOS simulator, Android emulator, or Android device from macOS through PhoneAgent's JSON-RPC bridge.

    798 GitHub stars~2.2k tokensUpdated 1 mo ago
    MobileAuto-check passed

More from lycorp-jp/sim-use

  • Bump Version Dev

    lycorp-jp/sim-use

    Build a release-shaped sim-use binary and install it locally for testing.

    1.4k GitHub stars~1k tokensUpdated yesterday
    Auto-check passed
  • Release

    lycorp-jp/sim-use

    Cut a sim-use release end-to-end. An agent skill from lycorp-jp/sim-use.

    1.4k GitHub stars~1.7k tokensUpdated yesterday
    Auto-check passed
  • Run Evals

    lycorp-jp/sim-use

    Prepare the environment and run the LLM-driven agent evals (e2e/agent-evals/) against a chosen sim-use binary.

    1.4k GitHub stars~1.4k tokensUpdated yesterday
    Auto-check passed

Works with

Categories

Questions about Sim Use

What does Sim Use do?

Drive iOS Simulator, Android emulator/device, and physical iPhone/iPad screens for AI agents. Sim Use is an agent skill from lycorp-jp/sim-use. Drive iOS Simulator, Android emulator/device, and physical iPhone/iPad screens for AI agents.

When should I use Sim Use?

Sim Use fits situations like: asked to automate a simulator; drive a real iOS device; tap/swipe/type on a device; take a screenshot.

How do I install Sim Use in Claude Code?

Run `npx skills add lycorp-jp/sim-use --skill sim-use -a claude-code`. Or copy the skill folder (skills/sim-use in lycorp-jp/sim-use) into .claude/skills/sim-use in your project. Claude Code loads it when a task matches its description.

How do I install Sim Use in Codex?

Run `npx skills add lycorp-jp/sim-use --skill sim-use -a codex`. Or copy the skill folder (skills/sim-use in lycorp-jp/sim-use) into .agents/skills/sim-use in your project. Codex loads it when a task matches its description.

Can I use Sim Use in Cursor, Gemini CLI or GitHub Copilot?

Cursor, Gemini CLI, GitHub Copilot and OpenCode also load SKILL.md folders. With the skills CLI, run `npx skills add lycorp-jp/sim-use --skill sim-use -a cursor` (or -a gemini-cli, github-copilot or opencode for the others). To copy it by hand, put the folder in .cursor/skills/sim-use, .gemini/skills/sim-use, .github/skills/sim-use and .opencode/skills/sim-use in your project.

What does Sim Use need to run?

Going by SKILL.md and its folder, Sim Use needs Python for the scripts in its folder and the command-line tools its instructions call (python3). Our summary lists: Python 3.

Does Sim Use access the network?

SKILL.md contains no URLs. Any network use would come from the scripts or tools the agent runs. This is read from the text; nothing was executed.

Is Sim Use safe to install?

Our automated static check of SKILL.md found no risky patterns, such as piping downloads into a shell, reading credential files or hidden Unicode. It is not a guarantee. The check reads SKILL.md only: the scripts in the folder are not scanned, so read them before running anything.

What licence does Sim Use use?

Sim Use is published under the Apache-2.0 licence (the repository's licence). It allows redistribution, so the full SKILL.md is shown on this page.

How many tokens does Sim Use use?

About 3.2k tokens (SKILL.md is roughly 13k characters). Agents keep only the skill's name and description in context until a task matches; then they load SKILL.md in full. Its references folder adds about 5.5k tokens, read only when the agent opens those files.

What are the alternatives to Sim Use?

Skills that share tags, products or a category with Sim Use: Mobilerun Docs Reference (droidrun/mobilerun, 9.6k stars), App Store Screenshots Generator (ParthJadhav/app-store-screenshots, 7.2k stars), Rnn Codebase (wix/react-native-navigation, 13k stars) and Building Native UI (CherryHQ/cherry-studio-app, 4k stars). The comparison table on this page puts their stars, adoption, token cost, safety result and licence side by side.

Who maintains Sim Use?

lycorp-jp (a GitHub organization) maintains it in lycorp-jp/sim-use, which has 1,393 GitHub stars. The repository holds 4 skills in this directory. The repository was last updated on October 6, 2026.

Source: lycorp-jp/sim-use on GitHub. Facts on this page come from the repository at the commit we read; the author's words are quoted as theirs.