The Best Testing and TDD Skills for Claude Code

"Testing" means two unrelated things in a skill directory: software tests and marketing experiments. This post covers the first. Installing a marketing A/B test skill when you wanted Playwright is a real mistake, and the descriptions do not always make the distinction obvious.

Five skills in our directory address software testing, and they score within a very narrow band — 9.0 to 9.1. That uniformity is informative: there is broad agreement on what good testing practice looks like, and the differences are in scope rather than quality.

The skills worth installing

test-driven-development — 9.1 · Developer Tools

The red-green-refactor loop: write the failing test first, then the implementation. What matters here is sequencing — an agent told to "add tests" will usually write tests that describe the code it just wrote, which is a different and weaker artifact.

replica-test — 9.1 · Developer Tools

Clicks through every flow of an app and tests it for bugs, producing a test plan and results. Aimed at replication projects, but the flow-based approach is useful for any app where the tests should follow user journeys rather than functions.

playwright-skill — 9.0 · Developer Tools

Browser automation with Playwright: auto-detects dev servers, writes and runs tests. The infrastructure skill — useful when you need real browser behavior rather than a DOM simulation.

playwright-expert — 9.0 · Developer Tools

E2E test architecture: fixtures, page objects, parallelization, and test infrastructure. Overlaps with the one above; choose by whether you need to run a browser or structure a suite.

test-master — 9.0 · Developer Tools

Generates test files, designs mocking strategies, analyzes coverage. The generalist — the right first install if you have no existing testing setup.

The distinction that bites

The marketing category contains skills named message-test-designer, ad-test-designer, and email-quality-auditor. They are excellent and they have nothing to do with software tests. If your agent loads one of those when you asked for unit tests, the output will look plausible and be useless.

That is a trigger-boundary problem, and it is exactly what the trigger-quality dimension measures. It is also a reason to install only the testing skills you need rather than a broad collection — fewer overlapping triggers means fewer wrong routing decisions.

How to choose

• No tests at all? test-master, then test-driven-development.
• Browser flows? playwright-skill for running, playwright-expert for structure.
• Testing an existing app end to end? replica-test.

The developer tools category covers review, debugging, and architecture.