expo-skill-eval
An agent skill by expo, from expo/skills. Tags: analysis, evaluation, mobile, quality, screenshots, testing.
What it does
Evaluate Expo skills in this repo end-to-end - trigger accuracy, generated code quality, and runtime screenshots on iOS simulator and Android emulator via Expo Go (web optional). Use when the user wants to eval an Expo skill, test that a skill produces working code, benchmark a skill with device screenshots, or verify a skill's output renders correctly.
Install
With the skills CLI, which installs into Claude Code, Codex, Cursor and other agents:
npx skills add expo/skills --skill expo-skill-eval
Or copy the skill folder into Claude Code's skills directory by hand (~/.claude/skills for every project, or .claude/skills inside one):
git clone --depth 1 https://github.com/expo/skills
cp -r skills/.claude/skills/expo-skill-eval ~/.claude/skills/expo-skill-eval
Safety box score
Grade D, elevated risk: 61 of 100 risk points, rated 2026-09-28. The grade rates what the skill and its scripts can reach on the machine of whoever installs it, across eight categories from shell execution to secrets access. How the score works.
This skill is a full Expo eval harness: it creates fixture apps, runs `claude -p` subprocesses, launches iOS/Android/web snapshots, and can optionally publish an HTML/Artifact viewer. The main risks are broad code execution, runtime network downloads, and recursive cleanup of fixture state and caches; it also asks for `--dangerously-skip-permissions` on subprocesses.
| Category | Level | Why |
|---|---|---|
| Code & shell execution | 3 (high) | High: the harness deliberately executes shell/device commands and generated app code (`bunx expo start/run`, `adb`, `xcrun simctl`, `playwright`, and `claude -p` subprocesses), so it has broad local code-execution reach. |
| Network access | 3 (high) | High: it queries npm (`npm view expo dist-tags --json`) and downloads packages/browser binaries at runtime (`bunx create-expo-app`, `bunx expo install`, `bunx playwright install chromium`), so it both fetches and sends data over the network. |
| Filesystem reach | 3 (high) | High: it writes under `/private/tmp/expo-skill-eval-*`, caches fixtures in `~/.cache/expo-skill-eval/fixtures`, and recursively deletes fixture trees and DerivedData with `rm -rf`. |
| Secrets & credentials | 0 (none) | No credential stores, `.env` files, keychains, or browser sessions are read or written; the only nearby env handling is stripping `CLAUDECODE`. |
| Privilege & persistence | 1 (low) | Low: it does not need sudo or system services, but it does require a global Claude plugin-state change via `/plugin` to disable the published `expo` plugin during trigger evals. |
| Instruction hijack surface | 2 (moderate) | Moderate: the workflow explicitly asks to confirm and then use `--dangerously-skip-permissions` for subprocesses, and it instructs the model to run those subprocesses unattended. |
| Irreversible actions | 2 (moderate) | Moderate: it boots simulators/emulators, installs/launches apps, captures screenshots, cleans fixtures, and can optionally publish a shareable Artifact, so it performs substantial autonomous actions beyond the workspace. |
| Transparency | 2 (moderate) | Moderate: the repo text is readable, but the exact code executed at runtime comes from unpinned `bunx`/registry downloads, so a reviewer cannot inspect the full execution set from the checked-in files alone. |
Before installing
- Read `scripts/clean-fixture.sh` and `scripts/make-fixture.sh`: they `rm -rf` fixture trees, `~/.cache/expo-skill-eval/fixtures`, and DerivedData.
- Decide whether you are comfortable with runtime downloads from npm/Playwright and with `--dangerously-skip-permissions` for `claude -p` subprocesses.
- If you plan to run trigger evals, be ready to disable the published `expo` plugin via `/plugin` first.
- Keep Artifact publishing off unless you want the final viewer to leave your machine.
Rated 2026-09-28 by an automated review of 13 files from the skill's repository.
Source
- Repository
- expo/skills (all skills from this repository)
- Path
- .claude/skills/expo-skill-eval/SKILL.md
- Branch
- main
- Updated
- 2026-09-22
Related skills
- eas-simulator — EAS service (paid). Run and control a user's app on a remote iOS/Android simulator hosted on EAS cloud.
- expo-dev-client — Framework (OSS). Build and distribute Expo development clients locally or via TestFlight for internal testing.
- expo-skill-feedback — Submit feedback on an Expo skill—or Expo itself—and control bundled anonymous usage telemetry (off by default / opt-in).
- ab-testing — When the user wants to plan, design, or implement an A/B test or experiment, or build a growth experimentation program.
- arena — Spawn N parallel candidates at the same task, pick a base, graft the strongest parts of the losers into it.
- audit-prep-assistant — Prepares codebases for security review using Trail of Bits' checklist.