- Home
- Skills
- Testing & Debugging
- qa-pilot — QA for every web development task
Works with the AI Agents you already use
qa-pilot — QA for every web development task
Checks acceptance criteria before you code, verifies the diff after, and writes Playwright and unit tests in your project's own conventions, each with proof that it fails when the code breaks.
Free
About this skill
qa-pilot does the QA step of a web development task inside your coding agent. It has two commands: onboard once per project, then test on every task.
onboard reads your stack, test frameworks, conventions and CI and writes docs/qa/project-profile.md. Every later run follows that profile. qa-pilot uses your existing setup and does not bring its own framework. During setup it also compares your test layers with the code and notes the gaps, such as logic without unit tests or layers that run only locally, in docs/qa/strategy-check.md.
test is the everyday command. Give it a ticket, a branch or a URL, or just ask in plain words, and it picks the work:
- Before coding: checks the acceptance criteria against the task logic and the existing system, maps what the change reaches, and sets which tests to change or add, sorted into Must / Should / Highlight.
- After coding: reconciles the diff with that plan, asks "intended?" about behaviour changes nobody planned, proves coverage, and writes only the missing tests, under strict limits.
- For a feature: a prioritised test plan, and the tests themselves in Playwright and your lower-level frameworks.
- For code that already shipped: the tests it never got.
- In the browser: exploratory testing of a flow in a real browser.
Every new test goes through a failure probe. The probe breaks the code under test, checks that the test goes red, and restores the code. A test that cannot fail is reported, not shipped.
What's included
- Two commands:
onboardandtest. - 80+ QA rules: web failure patterns (modify vs delete, double submit, silent mutations, UI-only permissions…), web invariants, test strategy, test-code quality and UI testing.
- Two stdlib Python scripts: the failure probe and a profile drift check.
Requirements
A coding agent with terminal access, Git and Python 3. For browser testing, a Playwright MCP server (the pack prints the config). Skill texts are in English; the rule texts are in Russian.
qa-pilot reports bugs and does not change product logic. The one exception: during browser testing, on your request, it fixes small, low-risk frontend bugs (typo, broken handler, style). Auth, data and backend issues are only reported.
Audits are in qa-pilot Pro
Whole project, a feature's requirements, the test suite, a UI component library: qa-pilot Pro.
Licence
Free for any use, business included; no redistribution.
One development task at a time; audits of a whole project, a feature's requirements, the test suite or a UI library are in qa-pilot Pro. Reports bugs; does not change product logic, except small low-risk frontend fixes during browser testing when you ask. Needs a one-time onboard run. Built for web projects. Rule texts are in Russian.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
This skill has been used 3 times.
Trust & safety
Security scan
Security scan passed 17 days ago
- Free with an Agensi account
Requires
Permissions
Terminal: runs your project's tests and the pack's stdlib Python scripts. Files: reads the project; writes tests, e2e config and fixtures, test ids in components, and reports under docs/qa/; small frontend fixes only on request during browser testing. Browser: only browser testing, through Playwright MCP. Network: only when you approve installing test tools (e.g. Playwright) during e2e setup, and the browser opens your app's URL.