← ClaudeAtlas

test-designlisted

Use when a spec is finalized and tickets are about to be cut, or when loose tickets with acceptance criteria and no spec are about to be dispatched. Writes docs/qa/plan-<slug>.md, the risk-ranked plan of journeys, endpoints and jobs that QA later runs against, with the cases each ticket must carry. Plans only; writes no tests and cuts no tickets.
thientranhung/astragentic · ★ 4 · Testing & QA · score 70
Install: claude install-skill thientranhung/astragentic
# Test design: the plan exists before the tickets do **A plan written after the build lists what was built. A plan written before it lists what could go wrong.** QA's run reads this file as its oracle, so a journey, endpoint or job missing here is one nobody exercises. Without it QA derives the oracle from the source and the verdict loses confidence: an inferred oracle never earns an unconditional pass. ## When it runs In the Shaper session, after `astra-to-spec` and the spec arm, before `astra-to-tickets`; and, since no ticket is dispatched without a plan, over a set of loose tickets that have no spec. Two inputs, one of them required: - **A committed spec** with its acceptance criteria: the plan covers it whole (one unit, whether the spec is a system or a single epic). - **Loose tickets with no spec** (a gate's follow-ups, small fixes): the input is each ticket's acceptance criteria plus the source it cites (a gate report, a defect record). The plan has one section per ticket, and a ticket with no surface gets its line: `QA: none — <why>`. An input with no stated way to know it works — a spec without criteria, a ticket with none — halts the skill. Name what is missing and hand back. ## Method 1. **Detect the surfaces.** Frontend, backend or fullstack from the manifests. List what the spec adds or changes by kind: screens and journeys, endpoints, jobs and data flows. Each kind selects an execution mode for QA: walk, probe or verify. 2. **Read what already