Imported from htest-dev/htest (SKILL.md). Install upstream with npx skills add htest-dev/htest. Copyright stays with the author.

hTest Tests

Two modes: JS-first for logic (runs in Node or browser, CI-compatible) and HTML-first for UI (runs in browser only, reactive).

Full reference: https://htest.dev. This skill summarizes common patterns and pitfalls — when in doubt about a property's exact semantics, the docs are the source of truth.

JS-First Mode

Tests are nested object literals; properties cascade from parent to child, so you only specify what differs.

Core Pattern

import { myFn } from "../src/my-fn.js";

export default {
	name: "myFn()",
	run: myFn,
	tests: [
		{
			name: "descriptive name",
			arg: "input",
			expect: "output",
		},
	],
};

Best Practices

Two principles guide everything below. Come back to these whenever a specific rule seems wrong for your test:

• What abstraction makes the tests easier to understand for humans? If a rule below would impose an abstraction that makes a test harder to read, the rule is wrong for that test. Declarativeness exists to remove repetitive noise — not to force inherently imperative code into a data shape.

• Write the tests first; scaffold around them. Express each test as data (name, arg, expect) in whatever shape gives you maximum signal-to-noise on the intent. DRY isn't the endgame — it often correlates with clarity, but not always. Write run(), beforeEach(), and other scaffolding at the end, derived from what the tests need. When deciding whether to factor run() out or leave it per-test, write the test both ways and pick the version that expresses intent better.

1. arg + expect is the default for value transforms. When a test naturally fits "given input, expect output," express it that way. Don't bend stateful or imperative tests into this shape. Lifting run to a shared level pays off only when the abstraction simplifies. If it would require encoding sequential side effects as arg data — an array of functions whose job is to mutate state — leave run per-test: the imperative content stays where the reader expects it. When several tests genuinely share execution, group those under a focused run rather than forcing the whole file under one root-level shape.
2. Structure mirrors API — nest test groups to match your module's shape. One export = one top-level group. The hierarchy should let you locate any test by navigating the API.
3. Check the shape, not every byte — use check: { subset: true } when you only care about specific properties. Tests that over-specify expect break for wrong reasons.
4. Each test = one unique branch — don't test the same code path twice with different data. A passing duplicate isn't a safety net.
5. Don't test the language — spread, object merge, array methods: those are JavaScript, not your code. Only test behavior your function owns.

Common Mistakes

Test Object Properties

All properties are optional and inherit from parent to child.

Execution

Structure

skip: "fail" runs the test but treats failure as skipped (the result is still shown). A passing test counts as a regular pass. Useful for marking nice-to-have tests — progress is visible without breaking CI.

Comparison

Default check is deep(shallowEquals()) — recursive === at leaf level.

check Object Shorthand

Pass an object instead of a function to configure built-in comparison behavior:

subset: true is the most useful option. Use it when results may contain extra fields you don't want to validate:

{
	run: parseItem,
	check: { subset: true },   // inherited by all children
	tests: [
		{ arg: "foo:bar", expect: { key: "foo", value: "bar" } },
		// passes even if parseItem() also returns { raw: "foo:bar", index: 0, ... }
	],
}

Combine options freely:

check: { epsilon: 0.005 }                    // numeric tolerance, deep (default)
check: { subset: true, epsilon: 0.0001 }     // partial match + tolerance
check: { deep: false, looseTypes: true }     // shallow loose equality

For a custom comparison, use an inline function — no import needed:

{
	check: (actual, expect) => Math.abs(actual - expect) < 0.01,
	tests: [...],
}

For pre-built utilities and composition, import the check module:

import check from "htest.dev/check";

export default {
	run: computeValue,
	check: check.proximity({ epsilon: 0.005 }),          // pre-built deep + tolerance
	// check: check.deep(check.proximity({ epsilon: 0.005 })),  // composed
	tests: [...],
};

Lifecycle Hooks

A child that defines its own beforeEach/afterEach overrides the parent's — they are not chained automatically.

To invoke the parent's hook from a child override, call this.parent.beforeEach() (or whichever hook). You control where the parent's logic runs — before, after, or in the middle of the child's:

{
	beforeEach () {
		this.parent.beforeEach();   // parent's setup first
		this.data.extra = "child";  // then child-specific setup
	},
}

Lifecycle hook errors

If a hook throws, the test is skipped (not failed). Hooks are infrastructure — if setup fails, the test result would be meaningless; if cleanup fails, the test environment is unreliable. A hook error never fulfills throws.

this Inside run

run is called with this set to the Test instance. Available properties:

• this.args — argument array
• this.data — inherited data object
• this.name — test name
• this.level — nesting depth (root = 0). Useful in getName for depth-aware labels
• this.parent — parent test/group. Useful for extending the parent's run in a child: call this.parent.run(...args) first, then transform the result
• this.expect — expected value

Async

hTest auto-awaits Promises returned from run. No special handling needed:

{
	async run () {
		let result = await fetchSomething();
		return result.value;
	},
	expect: 42,
}

Parallel Execution

Tests at the same nesting level run in parallel. Don't rely on execution order or shared mutable state between sibling tests:

let counter = 0;

// ❌ Sibling tests run in parallel — counter increments are nondeterministic
{
	tests: [
		{ run: () => ++counter, expect: 1 },
		{ run: () => ++counter, expect: 2 },
	],
}

// ✅ Each test is self-contained
{
	tests: [
		{ run: () => "foo".toUpperCase(), expect: "FOO" },
		{ run: () => "bar".toUpperCase(), expect: "BAR" },
	],
}

Inheritance

Properties cascade from parent to child. Children only specify what differs:

{
	run: parse,             // Shared by all children
	tests: [
		{ arg: "foo", expect: { type: "foo" } },               // inherits run
		{ arg: "bar baz", expect: { type: "bar", mod: "baz" } },
	],
}

Override run at group level when a subset needs different logic:

let item = { id: 1 };

{
	name: "Promise",
	async run (arg) {
		let result = transform(Promise.resolve(arg));
		return { value: await result.value };
	},
	tests: [
		{ arg: item, expect: { value: item } },     // inherits async run
		{ arg: "foo", expect: { value: "bar" } },
	],
}

Reference Equality for Instances

hTest's default check uses === at leaf level. Share the same instance between arg and expect:

let config = new Map([["key", "value"]]);

// ✅ Same instance — passes
{ arg: config, expect: { result: config } }

// ❌ Different instances — fails
{ arg: new Map([["key", "value"]]), expect: { result: new Map([["key", "value"]]) } }

map — Both Sides

map transforms both result and expected before comparison. When either is an array, map is applied element-wise. Design so expected values pass through unchanged:

{
	map (result) {
		if (result?.items instanceof Set) {
			return { ...result, items: [...result.items] };
		}
		return result;
	},
	tests: [
		{ arg: { items: new Set(["a", "b"]) }, expect: { items: ["a", "b"] } },
	],
}

data — Shared Fixtures

data is a cascading object accessible to run and any setup hook via this.data. A child's data inherits from its parent's via the prototype chain, so properties set on a parent — even after construction (e.g., in beforeAll) — are visible to children. A child's own data properties shadow the parent's. The full definition lives in the docs (define / data) — what follows is patterns that come up in practice.

Good use: shared object, accessed via this.data

export default {
	data: {
		user: new User({ name: "alice", role: "admin" }),
	},
	run(method) {
		return this.data.user[method]();
	},
	tests: [
		{ arg: "isAdmin", expect: true },
		{ arg: "getLabel", expect: "Admin: alice" },
	],
};

Good use: nested groups vary one config value

export default {
	run(arg) {
		return transform(arg, this.data.mode);
	},
	tests: [
		{
			name: "strict mode",
			data: { mode: "strict" },
			tests: [{ arg: "foo", expect: "FOO" }],
		},
		{
			name: "loose mode",
			data: { mode: "loose" },
			tests: [{ arg: "foo", expect: "foo" }],
		},
	],
};

Good use: getData for fresh per-test data

When each test needs its own fresh data (e.g., an empty array to push into), use getData instead of beforeEach:

export default {
	getData () {
		return { items: [] };
	},
	run () {
		this.data.items.push(1);
		return this.data.items.length;
	},
	tests: [
		{ expect: 1 },
		{ expect: 1 },  // Each test gets its own fresh array
	],
};

Good use: setup hook builds a per-test fixture for run

When setup is more involved than a literal value — instantiating a class, opening a connection, assembling a DOM tree — do it in beforeEach and stash the result on this.data so run (and afterEach) can use it. beforeEach/afterEach receive the same args as run, so you can unpack them in the parameter list and write this.data.

export default {
	beforeEach ({ options }) {
		Object.assign(this.data, { parser: new Parser(options) });
	},
	afterEach () {
		this.data.parser.close();
	},
	run ({ input }) {
		return this.data.parser.parse(input);
	},
	tests: [
		{
			name: "strict mode rejects extra whitespace",
			arg: { options: { strict: true }, input: "1, 2, 3" },
			throws: SyntaxError,
		},
		{
			name: "loose mode parses anyway",
			arg: { options: { strict: false }, input: "1, 2, 3" },
			expect: [1, 2, 3],
		},
	],
};

Don't use data when the value is only used in one place — inline it instead:

// ❌ Unnecessary indirection
data: { prefix: "https://example.com/" },
run (arg) { return fn(arg, this.data.prefix); },

// ✅ Just inline it
run (arg) { return fn(arg, "https://example.com/"); },

Module-Level Setup

arg and expect expressions are evaluated at import time — during Test construction, before any hooks run. A variable that will be assigned in beforeAll is still undefined when arg captures it:

// ❌ db is undefined at import time — beforeAll hasn't run yet
let db;

export default {
	beforeAll () { db = createFakeDatabase(); },
	tests: [
		{ arg: db, expect: { rows: 0 } },  // arg captures undefined, not the db object
	],
};

// ✅ Module top level — exists when arg expressions evaluate
let db = createFakeDatabase();

export default {
	run: query,
	tests: [
		{ arg: db, expect: { rows: 0 } },
	],
};

This applies to any setup that arg/expect depends on: mock classes, shared fixtures, or preconfigured instances.

Data-Driven Test Generation

Test arrays are plain JS — you can build them with .map(), .flatMap(), or any array method before exporting:

const cases = [
	{ input: "foo", expected: "FOO" },
	{ input: "bar", expected: "BAR" },
];

export default {
	run: toUpperCase,
	tests: cases.map(({ input, expected }) => ({
		arg: input,
		expect: expected,
	})),
};

Useful for tests driven by an external data file, a registry, or computed expected values:

// Compute expected from a reference implementation
tests: rawCases.map(c => ({ ...c, expect: referenceImpl(c.arg) })),

Configuration

Optional. hTest looks for {,_,.}htest.{json,config.json,config.js} in the project root, or use --config <path> to load one explicitly. Every CLI flag maps 1:1 to a config key of the same name; CLI flags override config values.

Config keys: location, path, only, ci, watch, verbose, format ("rich" or "plain"), env ("auto", "node", "console"), setup.

setup — Pre-Test Scripts

Import scripts before any test file is loaded. Runs once, sequentially. Paths resolve relative to cwd. Each entry is a string or { src, loadIf } — loadIf is evaluated at config import time; if false, the script is skipped.

// htest.config.js
export default {
	setup: [
		"test/polyfills/dom.js",
		{
			src: "test/polyfills/structured-clone.js",
			loadIf: !globalThis.structuredClone,
		},
	],
};

Running Tests

npx htest test/file.js                  # Single file
npx htest test/                         # All JS in directory (not recursive, skips index*)
npx htest test/index.js                 # Use index files for recursive aggregation
npx htest test/file.js 0/1              # Run only the test at this position (positional)
npx htest test/file.js --only myTestId  # Run only the test with this id (or position); pass multiple times to combine
npx htest test/file.js --ci             # Force non-interactive mode (automatic in non-TTY environments)
npx htest test/file.js --verbose        # Show all tests, including passing (-V short)
npx htest test/file.js --watch          # Re-run automatically on file changes (-w short)
npx htest test/file.js --format plain   # Plain output instead of rich (-f short)
npx htest test/file.js --env node       # Force a specific environment (-e short)
npx htest test/file.js --config ./my.config.js  # Load a specific config file (-c short)
npx htest test/file.js --setup ./hooks.js       # Run a setup script first; pass multiple times to run several (-s short)
npx htest --help                        # List every flag (-h short)
npx htest --version                     # Print version (-v short)

HTML-First Mode

Use for UI-heavy code that needs a real DOM — web components, layout, interaction. Tests run in the browser only (not CI-compatible). Can be mixed with JS-first tests in the same suite.

Setup

Include hTest in an HTML file — no build step needed:

<link rel="stylesheet" href="https://htest.dev/htest.css" crossorigin />
<script src="https://htest.dev/htest.js" crossorigin></script>

Open the file directly in a browser. hTest bootstraps on DOMContentLoaded.

Reftests — Core Pattern

Each <table class="reftest"> contains tests as <tr> rows. The last two cells are compared (output | expected). When they match → pass; when they don't → fail.

<section>
	<h1>My Component</h1>
	<table class="reftest">
		<tr>
			<td><my-component value="42"></my-component></td>
			<td>42</td>
		</tr>
	</table>
</section>

3-column tables add a data/setup column before output and expected (first column is ignored in matching):

<tr>
	<td><!-- setup / ignored --></td>
	<td><!-- output --></td>
	<td><!-- expected --></td>
</tr>

data-test — Comparator

Default is "contents" — compares trimmed text content of the last two cells.

Can also be a JS expression or global function name receiving the cells as arguments.

data-click — Simulate Interaction

Automates clicks for reactive tests. Syntax: [selector] [wait Ns] [after eventname] [N times]

<!-- Click the button, then evaluate tests -->
<table class="reftest" data-click="button">
	<!-- Click after load, with delay -->
	<tr data-click="button wait 1s after load"></tr>
</table>

data-error — Expected Throws

<tr data-error>
	<td>
		<script>
			throw new TypeError("bad");
		</script>
	</td>
	<td>TypeError</td>
</tr>

Isolation / Debugging

• Click a <section> heading link → show only that section's tests
• Alt + double-click a <tr> → isolate that single test
• "Show all tests" link appears to restore the full suite

When to Use Each Mode