Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Expo's official example projects — the expo/examples repo of ~70 `with-*` integrations (Stripe, Clerk, Supabase, OpenAI, maps, Reanimated, SQLite, Skia, NativeWind, and more). Use when integrating a third-party library or service into an existing Expo app and you want the canonical,...
.claude/skills/sickn33-expo-examples/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | 92% | 0% |
| case-06 | ✗→✓ | ▲ Improved | 18% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 134% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 38% | 0% |
| case-12 | ✗→✓ | ▲ Improved | 34% | 0% |
Use this skill when you need expo's official example projects — the expo/examples repo of ~70 with-* integrations (Stripe, Clerk, Supabase, OpenAI, maps, Reanimated, SQLite, Skia, NativeWind, and more). Use when integrating a third-party library or service into an existing Expo app and you want the canonical,...
expo/examples is Expo's official library of ~70 integration examples — directories named with-<library> (e.g. with-stripe, with-maps), each built around one library or service. These are not full apps: they're managed projects (no ios//android/ dirs — native setup is via config plugins), and the typical one is a single screen of ~100–200 lines. Mine them for the canonical integration pattern — the dependency set, app.json config plugins, and minimal wiring Expo maintains against the current SDK — and adapt that into the user's app. Don't expect to lift an application architecture from them.
Reach for an example before hand-rolling an integration. (Kinds — full-stack, showcases, starters — are noted in ./references/catalog.md.)
Map the user's need to an example name (e.g. payments → with-stripe, auth → with-clerk). ./references/catalog.md is a categorized snapshot for fast triage — but it drifts, so confirm against the live list:
bash# Live example names: gh api repos/expo/examples/contents --jq '.[] | select(.type=="dir" and (.name|startswith(".")|not)) | .name' # Aliases (renamed) + deprecated (dead/moved) examples — check before recommending: gh api repos/expo/examples/contents/meta.json --jq '.content' | base64 -d
meta.json is the source of truth for what's renamed or dead (deprecated examples are removed from the repo tree but still listed here, each with a message). If an example is in its deprecated map, don't recommend it — follow the message to the modern path. If it's in aliases, use the destination.
The common case: the user already has an app and wants to see how Expo does something. Read the example as reference and apply the patterns by hand — never scaffold an example on top of their project.
First, list the whole example in one call. Integration code is often nested (e.g. Stripe's server routes live in app/api/), so a one-level listing misses the important files:
bashgh api 'repos/expo/examples/git/trees/master?recursive=1' \ --jq '.tree[].path | select(startswith("with-stripe/"))'
Then read the high-signal files first: README.md (setup) → package.json (deps) → app.json (config plugins / permissions) → the integration code the manifest revealed → .env (required secrets). Per file:
bashgh api repos/expo/examples/contents/with-stripe/utils/stripe-server.ts --jq '.content' | base64 -d # No gh? Raw URL (branch is master): curl -s https://raw.githubusercontent.com/expo/examples/master/with-stripe/utils/stripe-server.ts
Reading more than a couple of files? Many integrations are spread across server routes, a client provider, and config (Stripe is). Skip the per-file calls — pull the whole example into a throwaway/gitignored dir (not the user's project) and read it freely with Grep/Read, then apply by hand:
bashnpx degit expo/examples/with-stripe /tmp/expo-ref/with-stripe # clean copy, no git history # fallback without degit (sparse-checkout, no full ~64 MB clone): git clone --depth 1 --filter=blob:none --sparse https://github.com/expo/examples.git /tmp/expo-ref/examples \ && (cd /tmp/expo-ref/examples && git sparse-checkout set with-stripe)
Read from there with Grep/Read; delete the scratch dir when done.
bashnpx create-expo --example with-stripe # short form: npx create-expo -e with-stripe bun create expo --example with-stripe # with bun
When the user already has an app, add only what the example introduces; never overwrite their setup.
package.json pins won't match an older project. Add only the missing deps with npx expo install <pkg> (it resolves SDK-correct versions) instead of copying exact versions.app.json/app.config.* plugins and permissions the example introduces that the user lacks — keep their existing config block intact..env shape — it holds placeholders, never working secrets.Done when the integration code is ported and every dependency, config plugin, permission, and env var it needs is accounted for in the user's app — not when it merely looks wired up.
master, not main (matters for raw URLs and sparse checkout).https://launch.expo.dev/?github=https://github.com/expo/examples/tree/master/<example>.expo-tailwind-setupbuilding-native-uiexpo-moduleupgrading-expo./references/catalog.md — categorized snapshot of the example library for fast triage.| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 18,493 | 5,326 | -71% | 1 | 1 | 0% | 3,778 | 1,925 | -49% | 0 | 0 | — |
case-02 | fail→pass | 8,409 | 7,866 | -6% | 1 | 1 | 0% | 1,636 | 3,148 | +92% | 0 | 0 | — |
case-03 | pass→pass | 14,145 | 11,976 | -15% | 1 | 1 | 0% | 2,464 | 3,894 | +58% | 0 | 0 | — |
case-04 | pass→pass | 22,422 | 19,477 | -13% | 1 | 1 | 0% | 4,342 | 5,194 | +20% | 0 | 0 | — |
case-05 | pass→fail | 13,449 | 26,187 | +95% | 1 | 1 | 0% | 2,545 | 1,915 | -25% | 0 | 0 | — |
case-06 | fail→pass | 17,174 | 4,951 | -71% | 1 | 1 | 0% | 2,074 | 2,437 | +18% | 0 | 0 | — |
case-07 | pass→pass | 8,850 | 3,914 | -56% | 1 | 1 | 0% | 1,496 | 2,315 | +55% | 0 | 0 | — |
case-08 | fail→pass | 5,186 | 1,680 | -68% | 1 | 1 | 0% | 807 | 1,891 | +134% | 0 | 0 | — |
case-09 | fail→pass | 9,407 | 3,176 | -66% | 1 | 1 | 0% | 1,571 | 2,166 | +38% | 0 | 0 | — |
case-10 | pass→pass | 9,244 | 3,927 | -58% | 1 | 1 | 0% | 1,562 | 2,264 | +45% | 0 | 0 | — |
case-11 | pass→pass | 12,403 | 8,517 | -31% | 1 | 1 | 0% | 2,062 | 3,040 | +47% | 0 | 0 | — |
case-12 | fail→pass | 9,817 | 2,780 | -72% | 1 | 1 | 0% | 1,501 | 2,011 | +34% | 0 | 0 | — |
case-13 | pass→pass | 6,893 | 3,648 | -47% | 1 | 1 | 0% | 1,181 | 2,184 | +85% | 0 | 0 | — |
case-14 | pass→pass | 9,889 | 6,311 | -36% | 1 | 1 | 0% | 1,668 | 2,597 | +56% | 0 | 0 | — |
case-15 | fail→pass | 5,860 | 5,887 | +0% | 1 | 1 | 0% | 952 | 1,974 | +107% | 0 | 0 | — |
case-16 | pass→pass | 13,344 | 5,825 | -56% | 1 | 1 | 0% | 2,243 | 2,590 | +15% | 0 | 0 | — |
case-17 | pass→pass | 6,348 | 2,415 | -62% | 1 | 1 | 0% | 1,114 | 2,019 | +81% | 0 | 0 | — |
case-18 | pass→pass | 9,508 | 8,322 | -12% | 1 | 1 | 0% | 1,710 | 3,013 | +76% | 0 | 0 | — |
case-19 | pass→pass | 15,411 | 7,296 | -53% | 1 | 1 | 0% | 2,294 | 2,773 | +21% | 0 | 0 | — |
case-20 | pass→pass | 5,531 | 4,005 | -28% | 1 | 1 | 0% | 916 | 2,154 | +135% | 0 | 0 | — |
case-21 | pass→pass | 11,695 | 7,940 | -32% | 1 | 1 | 0% | 1,751 | 2,798 | +60% | 0 | 0 | — |
case-22 | fail→pass | 15,298 | 7,516 | -51% | 1 | 1 | 0% | 2,021 | 2,758 | +36% | 0 | 0 | — |
case-23 | pass→pass | 7,095 | 2,348 | -67% | 1 | 1 | 0% | 1,169 | 1,995 | +71% | 0 | 0 | — |
case-24 | pass→pass | 5,223 | 1,724 | -67% | 1 | 1 | 0% | 890 | 1,850 | +108% | 0 | 0 | — |
case-25 | fail→pass | 10,632 | 1,843 | -83% | 1 | 1 | 0% | 1,610 | 1,873 | +16% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 25 cases were attempted, and 23 counted toward the lift figure. The other 2 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +28 percentage points is the difference between those two pass rates over the 23 comparable cases. 2 cases got worse with the skill loaded, and they are included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.