← ClaudeAtlas

unioss-test-evidencelisted

Use when the tester stage builds its test-case set — derives black-box cases mechanically from changes.md + spec ACs + scope.md across three layers, enforces the risk-gated exception taxonomy, per-case evidence, failure severity, and an honest PASS/PARTIAL/FAIL verdict.
ttncode/unioss-plugins · ★ 0 · Code & Development · score 70
Install: claude install-skill ttncode/unioss-plugins
# UNIOSS Test-Case Derivation & Evidence Contract ## Overview The tester derives its own complete case set **mechanically** — coverage never depends on the dispatch prompt happening to name a scenario. Then the evidence contract makes every claim traceable: no pass without this-run proof, no case silently dropped. **Core principle:** derive the full case table before driving anything; a case is either RAN with evidence or SKIPPED out loud. ## Black-box mandate You are the customer's last line of defence, not the author's second opinion. You test the running system, never the source. - **Never open source or test code** to decide what to test or whether it works. `changes.md` is read as a **list of affected surfaces** (screens, flows, tables) — not as an implementation to inspect. - **A green PHPUnit suite is not evidence.** Never cite `UT_*.txt`, a PHPUnit count, or the coder's fast-verify result on any row. Unit tests prove the code does what its author expected; they say nothing about what the end user sees. The coder owns them, and their result never raises or lowers your verdict. - **Nobody's claim substitutes for your observation** — not the plan's, not the reviewer's, not the coder's "this works now". ## Derivation — build the case table first ### Three sources, all mandatory 1. **`changes.md`** — every changed flow / screen / table → **≥1 Normal + the mandatory exception classes below**. Four changed surfaces can never yield two test cases. 2. **Spec acceptanc