HavenSkills

Claude Testing skills

Write tests, drive a browser against a running app and verify a change before calling it done.

Skills
6 skills
Full write-ups
2

How to choose

Start with the layer that is actually broken: missing tests, a flaky suite, or a claim of success nobody verified.

Filter

All skills in this category

webapp-testing

Drive a running web application in a browser and report the visible behaviour of a flow.

by anthropics Skill guide
npx skills add anthropics/skills --skill webapp-testing
Author
anthropics
Repo
GitHub
Write-up
Read →

playwright-best-practices

Selector and waiting strategy for Playwright suites that have become flaky.

by currents-dev
npx skills add currents-dev/playwright-best-practices-skill --skill playwright-best-practices
Author
currents-dev
Repo
GitHub

playwright-cli

Record a browser flow you performed by hand, then replay it as a script.

by microsoft
npx skills add microsoft/playwright-cli --skill playwright-cli
Author
microsoft
Repo
GitHub

agent-browser

Navigate a live page and read a snapshot of what is actually rendered.

by vercel-labs Skill guide
npx skills add vercel-labs/agent-browser --skill agent-browser
Author
vercel-labs
Repo
GitHub
Write-up
Read →

tdd

Work in failing-test, passing-test cycles instead of writing tests after the implementation.

by mattpocock
npx skills add mattpocock/skills --skill tdd
Author
mattpocock
Repo
GitHub

test-driven-development

A failing test first, then the change, as part of a workflow that ends in verification.

by obra
npx skills add obra/superpowers --skill test-driven-development
Author
obra
Repo
GitHub

What testing skills are for

A testing skill fixes the thing agents do most often wrong: they claim a change works without running anything. The skills on this page cover the three layers of that problem — writing the test, driving the browser to prove the flow, and refusing to call the work finished on stale output.

Which one you need

  • webapp-testing (anthropics) — exercises a running application in a browser and reports what it saw. It needs the app running; pointed at a stopped process it reports connection errors as findings.
  • playwright-best-practices (currents-dev) — the one to install when the suite exists and is flaky. It is about selectors and waiting strategy: wait for a stable condition, never a fixed number of seconds.
  • playwright-cli (microsoft) — record a flow you perform by hand, then replay it as a script. Record first, script second; a script written from a description is a guess.
  • Agent Browser from Vercel Labs inspects rendered UI and reports the live state you can use to write an assertion.
  • tdd (mattpocock) — failing test, then passing test. The failing run is the evidence the test measures something.
  • test-driven-development (obra) — the same discipline inside a workflow that ends in verification, which is what you want when several agents touch one repository.

Where these skills break in practice

  • Tests written after the code. They pass on the first run because they were written to match the implementation. Nothing is being verified.
  • Fixed sleeps. waitForTimeout(2000) works on your machine and fails in CI. Wait for state, not time.
  • Selectors tied to styling. A class name used by the test changes with the next redesign. Use roles, labels and test ids.
  • Success from an exit code. The command returned 0; the page still shows a validation error. Assert on content you can read.
  • Flaky tests marked as skipped. A skipped test is a decision to stop knowing. Either fix it or delete it.

The check that catches most of it

  1. Run the new test against the code before the fix. It must fail, and for the expected reason — not a syntax error.
  2. Run the suite twice. A test that passes once and fails once is flaky, and flaky tests stop being believed within a week.
  3. Read the assertion, not the test name. “Should work” proves nothing.
  4. Keep the run's output next to the diff; stale output from before the last edit is not evidence.

Related categories

  • Browser automation skills — inspect pages and replay browser actions in an authorized test environment.
  • Mobile skills — the same test question on device builds, where the emulator is the running application.
  • Next.js skills — the framework-level setup when the app under test is a Next project.
  • Superpowers — the verification gate that follows the test run.

Frequently asked questions

Do I need both webapp-testing and playwright-best-practices? Yes, for different jobs: the first drives a browser to check behaviour, the second keeps an existing suite from going flaky. Install the second when your CI is already lying to you.

Can it test without a running server? No. Every skill on this page needs the application up. Start it yourself and confirm the URL responds before the run.

Will the agent write the tests I am missing? It will write them from your description of the behaviour. The description is the work — a vague one produces tests that assert on rendering.

How many tests should a change come with? One failing test that proves the bug or the new behaviour, plus the existing suite still passing. More than that is usually scope creep.

What if a test is flaky? Fix the wait or the selector. Skipping it converts a known problem into an unknown one.