Human-Guided Agentic Test Engineering Pipeline

On my team, test planning, case design, and automation were three separate activities, usually picked up by different people at different times. Each hand-off started by rebuilding the same context: the feature, its epic, the repository, and the shape of the existing test suite. Coverage drifted in TestRail as similar cases got written twice, and automation built without that context turned brittle.

The workflow I built carries one Jira ticket through scenario design, TestRail coverage, and Playwright automation as a single reviewed path. It stops for a person before it changes coverage, writes code, or opens a pull request.

The workflow opens with /start-planning. Before it proposes anything, it reads the ticket in full: acceptance criteria, technical notes, attachments, linked references, the parent epic, and any sibling tickets already done or in review. From that context it drafts risk-tiered scenarios and returns them for review.

Once the scenarios are approved, /test-cases maps them onto the existing TestRail project. It updates the cases that already match, creates the folders and cases that are missing, and returns every affected link so the coverage can be checked before the feature ships.

After deployment, /automate takes the approved cases together with the repository context and writes POM-based tests, each routed to the UI, API, visual, or component suite that fits. It builds on the page objects and fixtures already in the codebase rather than adding parallel ones.

Each step produces an artifact that someone can review before the next step begins.