Free to browse, every day. Go Pro for unlimited installs and Airsc MCP.

See Pro
VISign in

Best Testing & debugging resources for vibe coders

Updated 19/10/2026

This guide is for people who build with AI agents and do not want to trust the result blindly. AI can write code that looks right but fails when you run it. These resources help you check the work, run tests in a real browser, and track down bugs.

Start with the beginner pick. Then add a browser testing tool or an error monitor when your app grows. Most picks here are labeled safe. Two have a caution label, and we say so below.

  1. PluginSafe

    dotpals

    Rikinshah787

    This plugin helps users see what their coding agent actually did, verifies test passes, and makes agents fix failing tests.

    36MIT

    #1 · Why we picked it

    Beginner level, safe label and a 98 quality score, the highest here; it can catch an agent faking a test pass and push it to fix failing tests.
  2. PluginSafe

    verify

    opslane

    This Claude Code plugin helps you verify agent-generated code by booting your application, driving changes against approved acceptance criteria, and providing a detailed report.

    115MIT

    #2 · Why we picked it

    A Claude Code plugin that boots your app and checks changes against acceptance criteria, then gives a report you can read as evidence.
  3. PluginSafe

    playwright-skill

    lackeyjb

    A skill that enables coding agents to write and execute custom Playwright scripts for complex web flows.

    3.2kMIT

    #3 · Why we picked it

    Lets the agent write and run Playwright scripts, and you can watch the browser session; good for quick end-to-end tests.
  4. SkillSafe

    SeleniumBase

    seleniumbase

    An MCP server and framework for browser automation, web scraping, and end-to-end testing with stealth capabilities.

    13kMIT

    #4 · Why we picked it

    Popular (13k+ stars), safe label, and covers UI tests through an agent; expect some setup since it is a full framework.
  5. MCP serverSafe

    httprunner

    httprunner

    HttpRunner is an open-source API/UI testing framework that uses LLM technology for natural language driven test scenarios, supporting multiple platforms and offering broad test capabilities.

    4.3kApache-2.0

    #5 · Why we picked it

    Turns plain-language descriptions into API tests and can also cover UI and load testing; safe label, intermediate level.
  6. SkillSafe

    webperf-snippets

    nucliweb

    A library of deterministic JavaScript snippets and agent skills for measuring and debugging web performance (LCP, CLS, INP) via AI agents.

    1.5kMIT

    #6 · Why we picked it

    Helps the agent measure and debug slow pages, including LCP, CLS and INP, using fixed scripts rather than guesses.
  7. PluginCaution

    toolkit

    getsentry

    Provides agentic tooling as an MCP server and Claude plugin to integrate Sentry's error monitoring and performance debugging features into coding agents.

    913

    #7 · Why we picked it

    Lets Claude Code look at real Sentry errors and suggest fixes; it has a caution label and needs a Sentry project to be useful, so review it first.
  8. MCP serverCaution

    vibetest-use

    browser-use

    This MCP server launches browser agents to perform automated QA testing on websites for UI bugs, broken links, and accessibility issues.

    833

    #8 · Why we picked it

    Runs browser agents to find UI bugs, broken links and accessibility issues on a local or live site; caution label, so read the setup notes before use.

How to choose

  • If you are new, start with dotpals. It checks what the agent really did without needing a test framework.
  • Want proof that a feature works? Use verify or playwright-skill to run the app in a real browser.
  • Testing an API instead of a screen? httprunner is the better fit.
  • Prefer the safe label. For caution items like getsentry-toolkit and vibetest-use, read what access they need first.
  • Skip the reverse-engineering and CTF tools unless that is your actual goal. They are advanced and not for everyday app bugs.

FAQ

How can I tell if my AI agent faked a passing test?

Use dotpals. It is built to show which files the agent changed and which commands it ran, and to verify that tests really pass.

Do I need to write test code myself?

Not always. Playwright-skill lets the agent write the scripts, and httprunner and agent-qa take plain-language descriptions. You still need to read the results.

What does the caution label mean?

It means the safety check found something worth a closer look. You can still use the resource, but review its permissions and setup first.

Which one helps with slow pages?

Webperf-snippets. It runs measurement scripts for Core Web Vitals such as LCP, CLS and INP, so the agent works from real numbers.