mabl skills teach an AI coding agent how to work with mabl. Installed in a coding agent such as Claude Code, Cursor, or GitHub Copilot, they give your agent the context it needs to plan and author mabl tests, design coverage for a feature, edit existing tests, and debug failures.
The skills use the mabl MCP to do their work. While you can use the MCP on its own, mabl skills provide your coding agent with another layer of context so it knows which skill applies to a request and when one should hand off to another.
What's included
- Project setup - reads your workspace, applications, environments, and credentials once, then saves them where your agent will find them again. Run this before the others.
- Test authoring - plans a single test conversationally, then builds it, and checks that what was built matches what you asked for.
- Coverage design - designs a set of tests for a feature area rather than one test, then hands each one to test authoring.
- Direct editing - changes an existing test's steps without opening your application in a browser. See direct editing.
- Debugging - investigates a failed run, reproduces it in a real browser, and confirms a fix.
How the skills work together
The skills are written to route work to each other, which is what makes them more than five separate tools:
- Setup comes first. Project setup points forward to authoring and coverage design, so your agent knows what to do once a project is configured.
- Coverage design delegates. Asked to cover a whole feature, it explores your application and then calls test authoring for each test it decides you need.
- Authoring hands off repairs. When authoring finishes and validation finds a gap it can state exactly, it passes that fix to direct editing rather than rebuilding the test.
- Editing routes back out. Asked for something it can't do without seeing the application, direct editing hands the work to the mabl agent. Asked about a failing test, it points at debugging instead.
In practice this means you can describe what you want at whatever level is natural — "cover this new page", "fix the assertion in this test", "why did this fail" — and your agent picks the right starting point.
Install
Install from github.com/mablhq/skills, which has current instructions for each supported agent. The plugin installs the skills and their MCP connections in one step.
The mabl CLI also installs a smaller set of skills, covering authoring and debugging, with mabl agent install. See author mabl tests from your AI coding agent and debug mabl tests from your AI coding agent.