write-a-test

Imported from slvcn-jsh/barion (.agents/skills/write-a-test/SKILL.md). Install upstream with npx skills add slvcn-jsh/barion --skill write-a-test. Copyright stays with the author.

Write a Test

0. Identity

• Role: Principal SDET (Software Development Engineer in Test). Writes bulletproof, deterministic tests across the testing pyramid (Unit, Integration, E2E), eliminating flakiness and ensuring total coverage by adhering to established reference guidelines.
• Authority: Owns the automated-test writing workflow only. Never applies manual-testing or QA-checklist patterns.
• Normative base: core/fellowship/merry.md, rules/engineering/architecture-boundaries.md, rules/common/code-style-standards.md, references/anti-patterns.md, context/core-domains/testing-strategy.md.

1. Intent (9 Dimensions)

2. Trigger Matrix

3. Execution Workflow

Step 1: Identify Layer

• Action: Determine the testing layer: fast Unit test (pure logic/components in isolation), Integration test (bounded contexts), or full E2E test (full browser stack).
• Input: Task.
• Stop Condition: None.
• Validation: Layer determined (Unit / Integration / E2E).

Step 2: Consult References (MANDATORY)

• Action: Read the relevant documentation in context/core-domains/testing-strategy.md and rules/engineering/testing-pyramid-standards.md. For E2E tests, rely on user-facing locators (getByRole, getByLabel). For Unit tests apply isolated mocking strategies.
• Input: Layer.
• Stop Condition: If testing principles were not consulted before writing code, stop and review them - alignment is mandatory before writing code.
• Validation: Testing strategy consulted.

Step 3: Draft the Test

• Action: Write the test code. Use Vitest for Unit/Integration and Playwright for E2E. Utilize the Page Object Model (POM) and explicit fixtures where applicable.
• Input: Reference + task.
• Stop Condition: If the draft uses CSS/XPath where a user-facing locator exists, stop and replace with getByRole/getByLabel per the priority order.
• Validation: Test drafted in the layer's strict convention.

Step 4: Enforce Determinism

• Action: Verify all UI assertions are await expect() to utilize auto-retry, and unit tests have zero side-effects.
• Input: Draft test.
• Stop Condition: If any arbitrary timeout or hardcoded wait state remains, stop and remove it.
• Validation: Deterministic; no flakiness sources.

4. Output Specification

Produce the test code using the strict framework conventions for the targeted layer.

Unit/Integration (Vitest):

import { describe, it, expect, vi } from "vitest";

describe("[Module Name]", () => {
  it("[Specific Scenario]", () => {
    // Isolated logic assertion
  });
});

E2E (Playwright):

import { test, expect } from "@playwright/test";

test.describe("[Feature Name]", () => {
  test("[Specific Scenario]", async ({ page }) => {
    await page.getByRole("button", { name: "Submit" }).click();
    await expect(page.getByText("Success")).toBeVisible();
  });
});

5. Validation Gate

• Testing layer identified (Unit / Integration / E2E).
• Matching reference consulted before writing code (mandatory).
• Strict framework convention used (Vitest / Playwright).
• User-facing locators throughout; no arbitrary timeouts or hardcoded waits.
• All UI assertions auto-retry via await expect().
• Unit tests have zero side-effects.

6. Anti-Triggers and Calibration

• Over-execution: Executing this skill for manual testing plans, load testing, or infrastructure/deployment checks outside of the application test runners.
• Under-execution: Skipping the skill for basic configuration changes - execute even for editing vitest.config.ts or playwright.config.ts.
• Calibration: Priority order for locators: role label text test ID CSS/XPath (last resort).

7. Anti-Pattern Compliance

8. Versioning & Changelog

• Version: 2.0.0
• Changelog:
  • 2.0.0 (2026-08-09) - Elevated to Tier 5. Added Identity, 9-Dimension Intent, Trigger Matrix, per-step Action/Input/Stop/Validation, Validation Gate, AP compliance map, Versioning, Portability Matrix.

9. Portability Matrix

10. Examples

Input: "Write a test for the login button."

Output: The agent determines the layer. If E2E, it generates a Playwright test using getByRole('button', { name: /log in/i }). If Unit, it generates a Vitest component test isolating the button logic.