← ClaudeAtlas

stage-testlisted

Test stage of a factory run — writes one end-user test per acceptance criterion plus unit tests for the invariants, and records the criterion-to-test table the story gate reads. Use after the plan stage of a story, when the orchestrator hands over a plan, or on "/stage-test". The tests must compile and be red before any production code is written.
domain-centric-development/dca-marketplace · ★ 0 · Testing & QA · score 72
Install: claude install-skill domain-centric-development/dca-marketplace
# Write the tests for one story Input: the story, `tasks/<story>/plan.md`, and the `## Qualities` and `## Look and feel` of the product scope where the plan says a criterion touches them. Nothing else. Open the files the plan's `## Files` names first — the pattern to mirror, the fixture to reuse; search only for the rest. Output: the test sources, plus `tasks/<story>/tests.md` with a `## Files` section naming the test files written. You write **no** production behaviour. ## Do 1. Read the plan's criteria and the test shape it chose for each. 2. Write **one end-user test per criterion** — for a scenario, the test follows its steps: `Given` is the arrangement, `When` the one action, `Then` and each `And` after it an assertion, with the scenario's own values — in the shape the plan named, using the test frameworks the project already has. It asserts the behaviour a user or a caller can observe, not the internals. **Change no build file**: adding a dependency, a plugin, a source set or a runner is a stack decision, not part of a story. If the planned shape is impossible with what is installed, write the tests file with a `## needs-human` section naming the missing runner and stop — do not install it, and do not silently drop to a unit test that asserts less. The section names a decision record (`decision: <story>-<nn>`), written to `.agents/factory/decisions/<story>-<nn>.md` with the front matter `id: <story>-<nn>`, `story:`, `stage: test`, `asked