Best testing skills for Claude Code and Cursor in 2026

The testing tag in this directory is noisy: our tagger matches the word "test" in descriptions, so skill-creator, caveman-commit and a marketing A/B skill all sit near the top by installs. We ignored the tag and asked one question of each skill: does it change what the agent does when a test fails or when it is about to claim success? Skills about software tests only; experiment design for marketing is a different subject.

Then we checked the repository's last push, the file list for scripts, and the frontmatter description, because a skill with an empty description never triggers on its own.

What it does: when a test fails or behaviour is wrong, the agent must reproduce it, form ranked hypotheses, and test them one at a time before proposing a change. About 247K installs, 281K stars, updated 2026-09-03, 11 files, contains scripts, 9.5 KB.

When to use it: every bug. The catch: it is part of the superpowers collection, which expects its sibling skills and works best installed as a set; on its own it still helps but references workflows you will not have. The folder includes scripts, so read them.

npx skills add obra/superpowers@systematic-debugging -g -y

What it does: before the agent says fixed, passing or done, it has to run the relevant command and show the output. No output, no claim. About 199K installs, one file, 3.6 KB, no scripts.

When to use it: always; it is the cheapest skill on this page. The catch: it triggers on the moment of completion, so on long tasks it can add a final test run you did not ask for. That is the point, but budget for it.

npx skills add obra/superpowers@verification-before-completion -g -y

What it does: gives the agent helper scripts to start a local app, drive it with Playwright, check the DOM and capture screenshots and console output. About 149K installs, 173.6K stars, updated 2026-09-03, 6 files, contains scripts, 3.9 KB.

When to use it: checking that a UI change actually works, not just compiles. The catch: the scripts are Python and need Playwright installed with browsers downloaded; the first run on a clean machine fails until you do that.

npx skills add anthropics/skills@webapp-testing -y

What it does: Vitest conventions from Anthony Fu: configuration, vi.mock patterns, fake timers, coverage, workspace setups. About 34K installs, 5.8K stars, 21 files, 3.7 KB entry.

When to use it: any project on Vitest, especially when the agent keeps writing Jest idioms. The catch: last updated 2026-06-23, the oldest on this list. Vitest moves fast; check the version the reference files assume.

npx skills add antfu/skills@vitest -y

What it does: tests in a real Chrome through the Chrome DevTools MCP server: read the console, inspect network requests, record a performance trace, then fix what it finds. About 28K installs, 92K stars, updated 2026-09-03, one file, 14.5 KB.

When to use it: performance and runtime errors that unit tests cannot see. The catch: it does nothing without the DevTools MCP server configured in your agent, and the skill does not install that for you.

npx skills add addyosmani/agent-skills@browser-testing-with-devtools -g -y

What it does: the structural rules for an end-to-end suite: page object model, fixtures, test isolation, retry policy and how to stop a flaky test from being retried into a pass. Playwright and Cypress. About 22K installs, updated 2026-09-01, 4 KB.

When to use it: setting up or cleaning up an E2E folder. The catch: it is short and framework-agnostic; it will not write the Playwright config for you, and it says nothing about component tests.

npx skills add wshobson/agents@e2e-testing-patterns -y

property-based-testing teaches the agent to state invariants and let Hypothesis, fast-check, proptest or jqwik generate the cases. About 5.3K installs, 7K stars, updated 2026-09-02, 9 files. Use it for parsers, serialisers and anything with an inverse. The catch: it is written by a security firm and leans toward finding crashes; for business logic you still need example tests.

tdd-workflow enforces red, green, refactor with coverage targets and a rule against writing implementation before a failing test exists. About 2.8K installs, 13 KB, updated 2026-09-03. Use it on a team that has decided to do TDD. The catch: the 247K stars belong to the ECC repository, not this file, and the install count is low; treat it as promising rather than proven.

npx skills add trailofbits/skills@property-based-testing -y
npx skills add affaan-m/ECC@tdd-workflow -y

skill-creator (about 371K installs) and caveman-commit (about 336K) lead the tag by installs but are about writing skills and commit messages; the tag is wrong. ab-testing from marketingskills (about 54K) designs marketing experiments. firecrawl-qa (about 31K) needs a Firecrawl API key. playwright-expert from Jeffallan (about 4.4K) overlaps e2e-testing-patterns and was last updated 2026-08-07. The language-specific ECC testing skills (python, golang, kotlin, cpp) are worth a look for that language, but two of them have an empty description in our crawl.

Give the agent a bug you already understand and watch the first three actions. With systematic-debugging loaded it should try to reproduce before editing; with verification-before-completion it should end by running the suite and pasting the result. If it goes straight to a fix, open the skill's description and compare it to how you phrased the request; the match is what loads the skill.