Setup
The primary way agents drive Peach is the peach CLI. Use the same
peach path a human uses after the standalone install; there is no separate
agent install.
From inside any agent (Claude Code, Cursor, Codex, or anything that can run
shell commands), peach describe,
peach do tap-text "Sign in", and the rest of the
CLI work against whatever sim is running.
Pick an integration surface
| your agent | use | why |
|---|---|---|
| runs shell commands (Claude Code, Cursor, Codex, CI) | the peach CLI | fastest, React Native-aware reads, auto-settle, claim leases |
| runs shell and works in this repo often | CLI + peach skill install | skills teach the supported workflows once, per agent home |
| browser-only tooling (Chrome MCP, DOM automation) | the accessibility DOM mirror | the sim exposes real <button>/<input> proxies, so browser tools drive it like a website |
All three drive the same simulator through the same gesture pipeline; they differ only in who is calling.
Start the app’s existing Metro, Expo, React Native, or One server yourself.
Peach does not own that process. Run peach open <port> from the app
project root so font, splash-screen, and project-configuration discovery has
the same context as the development server.
The simulator window
On Apple silicon macOS and Linux x64, peach open shows the app in the native
desktop app, which is faster than the browser. Setup installs it, and on a Mac
the first peach open installs it if it is missing.
Replace 8081 with the running app’s port. Linux desktop installation uses a
.deb and sudo apt-get install, and the app needs a desktop session.
peach desktop install --native on Linux installs the standalone headless host
instead. Intel Mac and Linux arm64 use the browser. To explicitly open a browser
window, run peach open 8081 --driver playwright --headed.
The optional background daemon keeps the bridge ready between commands and works with either window.
Verifying
With a sim open:
If list is empty, start a sim first (peach open <port> or your usual
dev flow), then retry.
If the scan or live app exposes a missing Peach compatibility seam, the agent
should explain the finding and offer to send it with peach support report.
Nothing is submitted until you explicitly agree. The command previews its
bounded attachment set before sending; see Reporting Issues.
Skills
We ship a handful of skill files that teach agents how to use Peach well: setup, debugging (including accessibility review), performance, visual review, and test authoring. The standalone CLI installs them into the standard agent skill homes:
Peach skills: peach-setup, peach-debug, peach-perf, peach-visual, and peach-test. Pick what matches your task,
then restart the agent session so it reloads the skill list.
What’s next
- +CLI Use: the day-to-day commands an agent runs.
- +Agent Devtool (Inspect): the in-sim panel where you talk to an agent and watch it work.