AI plugin

qa-pilot

A Claude Code plugin that turns testing into a repeatable process. The agent plans checks before the code, verifies the task after it, writes tests to the project's rules and runs audits, backing every conclusion with a link to the code or the requirement.

The plugin is open source, with the code on GitHub. It installs with two commands in Claude Code.

  • Claude Code
  • Skills and agents
  • Playwright MCP
  • Python

What's inside

01Project profile

The /qa-pilot:onboard command studies the stack, test layers, commands and CI once, asks about what isn't in the code — safe environments, critical flows — and writes a profile the other commands read.

02QA for every task

Before development, /qa-pilot:test drafts a test plan and looks for gaps in the requirements; after development, it verifies the task, walks the flows in a browser and writes tests from the plan.

03Audits

The /qa-pilot:audit command assesses the whole project, one feature's requirements or the test code: false-green and flaky tests, duplicates. The output is a prioritised work list.

04A test must be able to fail

Every new test is checked: the agent breaks the code under test, sees the failure and reverts the change. The expected result comes from the spec or the requirement, not from the code itself.

05Priority by risk

Findings are split into Must, Should and Highlight by their impact on the product, not by coverage percentage. Should is capped at eight, so the list can actually be done.

06Rule base and scripts

Over 1,300 testing rules checked against standards (ISTQB, OWASP, W3C, NIST). Scripts without an LLM pick the rules for the product, validate reports and check the profile against the repository.

Task

AI agents write tests fast, but those tests are often shallow, over-mocked and pass whatever the behaviour. We needed a process in which the agent tests like an experienced QA engineer and the result can be checked.

Result

The plugin is published on GitHub and on its own landing page. Plans, reports and tests are saved as files in the user's repository, and an interrupted session can be resumed.

Technologies

Claude Code · Skills and agents · Playwright MCP · Python

Discuss a similar project

Briefly describe your idea — we'll reply with options and timelines.

Discuss a project