uat-test-design

Run the Skill against the repository root. It executes an inventory script to measure documentation, E2E specs, unit tests, and screens, confirms scope and environment with you, then delegates extraction to parallel sub-agents that write L1, L2, and L3 item files directly.

It enforces reading budgets of 10,000 lines and 30 files per sub-agent, splits oversized feature groups before delegation, and marks low-priority sources as lookup-only. The parent agent processes only metadata summaries, never item bodies.

Yes, but with a caveat. If only reverse-generated or implementation-derived sources exist, it continues and records a fallback note in the index stating that expected results are based on current implementation, leaving business validity judgment to the acceptance team.

Yes. Each L2 item gets a coverage judgment (E2E, Unit, E2E+Unit, or none) based on test-name grep matching. Items fully covered by E2E with no production-environment value are separated into covered files, while Unit-only items stay in the UAT body.

It writes Markdown files into a date-stamped directory under docs/uat/: an index with count summaries and hierarchy-split tables, L1 scenarios, per-group L2 files, L3 non-functional items, and a gaps report. An xlsx conversion is available on request.

If information sources are almost entirely absent, the Skill stops and asks for documentation rather than inventing expected results. It also caps practical scope around 30 feature groups or 800 L2 items, proposing phased narrowing beyond that.