Best Claude Code Skills for Testing in 2026

agent-browser is a browser automation CLI built for AI agents. The skill itself is short. It's a discovery stub. It tells the agent to install the CLI with npm i -g agent-browser && agent-browser install, then load the current instructions from it. Those instructions ship with each CLI version, so they stay in step with the tool.

Testing is one of its named jobs. Its description lists "test this web app", exploratory testing, dogfooding, QA and bug hunts. The command agent-browser skills get dogfood loads a dedicated guide for that. Other guides cover Electron apps, Slack and cloud browsers.

Two things to know before you install. Its allowed-tools line pre-approves the agent-browser commands, so the agent runs them without asking. Its description also tells the agent to prefer agent-browser over any built-in browser tool. Because the real instructions come from the CLI, your Skillpack lockfile pins the stub, not the guides.

Matt Pocock's tdd skill is short and opinionated. It treats TDD as a red to green loop and spends most of its words on what makes a test worth keeping. Tests check behavior through public interfaces, so they survive a refactor.

Its key idea is the seam: the public boundary where a test observes behavior. Before writing any test, the agent writes down the seams it wants to test and confirms them with you. No test goes on an unconfirmed seam. It also names three anti-patterns. Implementation-coupled tests break on a refactor. Tautological tests recompute the expected value the way the code does. Horizontal slicing writes every test before any code.

The loop is strict. One seam, one test and one minimal implementation per cycle. Refactoring is left out of the loop and handed to a review step, which the skill links to the repository's code-review skill. The folder adds two guides, on tests and on mocking.

This skill comes from Superpowers, Jesse Vincent's collection of workflow skills. It's built around one "Iron Law": no production code without a failing test first. If the agent wrote code before the test, the skill tells it to delete that code and start over.

It walks through red, green and refactor, and makes the agent watch each test fail before writing code. Most of its length goes to the excuses an agent uses to skip a step. A table of common rationalizations answers each one, and a list of red flags tells the agent when to stop. The folder adds a guide on writing good tests.

Pair it with verification-before-completion from the same repository (253.2K installs). Its rule is "no completion claims without fresh verification evidence". The agent must run the command and read the output before it says anything passes.

This is Microsoft's official skill for Playwright CLI, a command-line version of Playwright made for coding agents. The project's README says CLI commands are more token-efficient than the Playwright MCP server. They don't push large tool schemas or page trees into the agent's context.

The agent takes a snapshot of the page and gets a short reference for each element. It then acts on them with commands like click, fill and type. The skill covers tabs, storage, network mocking, tracing and video. Its reference files include a guide to test generation (plan, generate, heal) and one on writing Playwright tests. Another covers attaching screenshots and videos to pull requests.

It needs Node.js 18 or newer and npm install -g @playwright/cli@latest. Like agent-browser, its allowed-tools line pre-approves playwright-cli and npx playwright commands.

webapp-testing is Anthropic's own skill for checking a web app that runs on your machine. The agent writes native Python Playwright scripts rather than calling a separate CLI. A short decision tree tells it what to do: read static HTML directly, or start the server and inspect the page first.

For dynamic apps, it follows a "reconnaissance-then-action" pattern. It waits for the network to go idle, takes a screenshot or reads the DOM, finds the right selectors, and only then acts. The skill warns that skipping the wait is a common mistake.

The folder ships a helper, scripts/with_server.py, that starts one or more dev servers, runs your script and shuts them down. The Skillpack CLI flags it as an executable before copying it. You need Python and Playwright installed.

Currents, which makes a dashboard for Playwright testing, publishes this skill as a free reference. It's the largest folder on this list, with more than 50 guides. They cover locators, assertions, fixtures, page objects, authentication, network mocking, flaky tests, CI providers and sharding. Framework guides cover React, Next.js, Vue and Angular.

It's activity-based. The SKILL.md maps tasks to guides, so the agent opens only the file it needs. A task like fixing a flaky test pulls in the debugging guides, not the CI ones. Testing patterns go beyond end-to-end tests: component, API, visual regression, accessibility, Electron and browser extensions.

Note the gap between its installs and its 411 repository stars. Most of its adoption seems to come from people installing it through skills.sh rather than starring the repo.

Addy Osmani's TDD skill covers the same red-green loop, then adds a test strategy around it. It starts by having the agent find the repository's own test commands, and never assume a default like npm test.

Its "Prove-It Pattern" is the part to steal. When a bug is reported, the agent doesn't start with the fix. It writes a test that reproduces the bug, watches it fail, fixes the code, then runs the full suite. The skill also describes a test pyramid with about 80% unit tests, and prefers real implementations over mocks.

For browser checks it points to Chrome DevTools MCP. The same repository has a separate browser-testing-with-devtools skill (58.4K installs) that requires that MCP server.

Anthony Fu generates this skill from the vitest-dev/vitest repository. Its GENERATION.md file records the exact source commit. The current version is based on Vitest 5.1.0-beta.1 and was generated on October 23, 2026.

It doesn't teach a method. It's a reference, so the agent writes tests with the real API rather than one it half remembers. The SKILL.md is a set of tables that point to about 20 reference files. They cover configuration, the test API, hooks, mocking, coverage, snapshots, filtering, concurrency, type testing and benchmarking.

It pairs well with one of the TDD skills above. The TDD skill decides when to write a test, and this one makes sure the Vitest code is right.