feat: Add pest-testing rules from Tim's tweet

.ai/pest/3/skill/pest-testing/SKILL.blade.php

@@ -100,9 +100,28 @@
 
 Pest 3 provides improved type coverage analysis. Run with `--type-coverage` flag.
 
+## Performance
+
+### Database
+
+- **Prefer `LazilyRefreshDatabase` over `RefreshDatabase`.** Tests that never touch the DB skip the migration/truncation cost entirely. Only use `RefreshDatabase` when the test specifically needs fresh migrations.
+- **Use `$factory->recycle($model)`** to create a shared parent once and thread it through nested factory calls, instead of each nested factory creating its own parent.
+- **Create only the models the test needs.** Audit factory usage — avoid nested `create()` calls that spawn extra rows the test never inspects.
+
+### Faking
+
+Fake any external interaction that is not the subject of the test:
+
+- `Notification::fake()`, `Mail::fake()`, `Bus::fake()`, `Event::fake()`, `Queue::fake()`.
+- If the test hits an HTTP endpoint that dispatches jobs, fake the queue unless the test is asserting job behavior — otherwise jobs execute inline and pull in everything they touch.
+
+For project-level environment and CI optimizations, see [reference/performance.md](reference/performance.md).
+
 ## Common Pitfalls
 
 - Not importing `use function Pest\Laravel\mock;` before using mock
 - Using `assertStatus(200)` instead of `assertSuccessful()`
 - Forgetting datasets for repetitive validation tests
 - Deleting tests without approval
+- Enabling `RefreshDatabase` on every test even when the test does not hit the DB
+- Adding `Queue::fake()` but still asserting side effects of the job (those will never happen)

.ai/pest/3/skill/pest-testing/reference/performance.md

@@ -0,0 +1,39 @@
+# Test Performance Reference
+
+Environment and tooling optimizations for faster Laravel test suites. These are project-level or CI-level settings — not per-test code changes.
+
+## Environment and config
+
+- **Set `BCRYPT_ROUNDS=4`** in `.env.testing` (or `phpunit.xml`). Default is 12 and hashing dominates auth tests.
+- **Disable XDebug.** Disable pcov too at scale unless you specifically need coverage.
+- **Disable background packages** in the testing environment: Pulse, Telescope, Nightwatch, and similar 3rd-party packages that do work on every request/command.
+- **Use `WithCachedConfig` and `WithCachedRoutes` traits** to avoid re-parsing config and routes on every test.
+- **Call `withoutVite()` (or `withoutMix()`)** in your test setup so the framework does not try to resolve built assets.
+- **`Http::preventingStrayRequests()`** in your base `Pest.php`. A single slow stray request can punish every test. Note this only catches requests made through Laravel's HTTP client. Audit direct Guzzle / cURL usage separately.
+- **`Sleep::fake(syncWithCarbon: true)`** in your base `Pest.php` so retries and backoffs do not actually sleep.
+- **`Exceptions::fake()`** to make sure you are not reporting to Flare/Sentry/Bugsnag from within tests.
+
+## Profile and diagnose
+
+- **`./vendor/bin/pest --profile`** prints the slowest tests per shard. Start there. Patterns in the top-10 often apply suite-wide.
+- When a slow test is mysterious, instrument event listeners or add temporary logging to find the unexpected work happening behind the scenes.
+
+## Quick reference
+
+| Area | Action |
+|------|--------|
+| Hashing | `BCRYPT_ROUNDS=4` |
+| Debuggers | Disable XDebug, disable pcov |
+| Background pkgs | Disable Pulse, Telescope, Nightwatch |
+| Config / routes | `WithCachedConfig`, `WithCachedRoutes` |
+| Assets | `withoutVite()` / `withoutMix()` |
+| HTTP | `Http::preventingStrayRequests()` |
+| Sleep | `Sleep::fake(syncWithCarbon: true)` |
+| Exceptions | `Exceptions::fake()` |
+| Profiling | `pest --profile` |
+
+## Common environment mistakes
+
+- Running the suite with XDebug loaded "just in case" a test fails.
+- Leaving `BCRYPT_ROUNDS` at the default because `.env.testing` was never created.
+- Calling real `sleep()` or using Carbon in production code, which prevents `Sleep::fake()` from helping.

.ai/pest/4/skill/pest-testing/SKILL.blade.php

@@ -146,10 +146,29 @@
     ->toHaveSuffix('Controller');
 @endboostsnippet
 
+## Performance
+
+### Database
+
+- **Prefer `LazilyRefreshDatabase` over `RefreshDatabase`.** Tests that never touch the DB skip the migration/truncation cost entirely. Only use `RefreshDatabase` when the test specifically needs fresh migrations.
+- **Use `$factory->recycle($model)`** to create a shared parent once and thread it through nested factory calls, instead of each nested factory creating its own parent.
+- **Create only the models the test needs.** Audit factory usage — avoid nested `create()` calls that spawn extra rows the test never inspects.
+
+### Faking
+
+Fake any external interaction that is not the subject of the test:
+
+- `Notification::fake()`, `Mail::fake()`, `Bus::fake()`, `Event::fake()`, `Queue::fake()`.
+- If the test hits an HTTP endpoint that dispatches jobs, fake the queue unless the test is asserting job behavior — otherwise jobs execute inline and pull in everything they touch.
+
+For project-level environment and CI optimizations, see [reference/performance.md](reference/performance.md).
+
 ## Common Pitfalls
 
 - Not importing `use function Pest\Laravel\mock;` before using mock
 - Using `assertStatus(200)` instead of `assertSuccessful()`
 - Forgetting datasets for repetitive validation tests
 - Deleting tests without approval
 - Forgetting `assertNoJavaScriptErrors()` in browser tests
+- Enabling `RefreshDatabase` on every test even when the test does not hit the DB
+- Adding `Queue::fake()` but still asserting side effects of the job (those will never happen)

.ai/pest/4/skill/pest-testing/reference/performance.md

@@ -0,0 +1,39 @@
+# Test Performance Reference
+
+Environment and tooling optimizations for faster Laravel test suites. These are project-level or CI-level settings — not per-test code changes.
+
+## Environment and config
+
+- **Set `BCRYPT_ROUNDS=4`** in `.env.testing` (or `phpunit.xml`). Default is 12 and hashing dominates auth tests.
+- **Disable XDebug.** Disable pcov too at scale unless you specifically need coverage.
+- **Disable background packages** in the testing environment: Pulse, Telescope, Nightwatch, and similar 3rd-party packages that do work on every request/command.
+- **Use `WithCachedConfig` and `WithCachedRoutes` traits** to avoid re-parsing config and routes on every test.
+- **Call `withoutVite()` (or `withoutMix()`)** in your test setup so the framework does not try to resolve built assets.
+- **`Http::preventingStrayRequests()`** in your base `Pest.php`. A single slow stray request can punish every test. Note this only catches requests made through Laravel's HTTP client. Audit direct Guzzle / cURL usage separately.
+- **`Sleep::fake(syncWithCarbon: true)`** in your base `Pest.php` so retries and backoffs do not actually sleep.
+- **`Exceptions::fake()`** to make sure you are not reporting to Flare/Sentry/Bugsnag from within tests.
+
+## Profile and diagnose
+
+- **`./vendor/bin/pest --profile`** prints the slowest tests per shard. Start there. Patterns in the top-10 often apply suite-wide.
+- When a slow test is mysterious, instrument event listeners or add temporary logging to find the unexpected work happening behind the scenes.
+
+## Quick reference
+
+| Area | Action |
+|------|--------|
+| Hashing | `BCRYPT_ROUNDS=4` |
+| Debuggers | Disable XDebug, disable pcov |
+| Background pkgs | Disable Pulse, Telescope, Nightwatch |
+| Config / routes | `WithCachedConfig`, `WithCachedRoutes` |
+| Assets | `withoutVite()` / `withoutMix()` |
+| HTTP | `Http::preventingStrayRequests()` |
+| Sleep | `Sleep::fake(syncWithCarbon: true)` |
+| Exceptions | `Exceptions::fake()` |
+| Profiling | `pest --profile` |
+
+## Common environment mistakes
+
+- Running the suite with XDebug loaded "just in case" a test fails.
+- Leaving `BCRYPT_ROUNDS` at the default because `.env.testing` was never created.
+- Calling real `sleep()` or using Carbon in production code, which prevents `Sleep::fake()` from helping.