What problem does it solve? Green test suites can hide vacuous assertions, non-deterministic tests, unbound fakes, and gates that cannot fail. This Skill judges tests that already exist — a directory, a file list, a diff, or "the tests for X" — and returns a severity-graded improvement plan instead of trusting a passing run. ## Core Features & Use Cases - Nine-pass review walk: Runs cheapest-falsifier-first passes from runner observability (Pass 0) through determinism, oracles, doubles, and red-proof verification (Pass 8), each with explicit stop conditions. - Shared ruleset with evidence: Cites T0/T1/T2 rules backed by proven failures in references/evidence.md, with stack- and kind-specific rule files loaded only when relevant. - Structured improvement plan: Produces a fixed-column findings table with work packages consumable by the /task skill; anything not executed is labelled "unverified (static reasoning)". - Use Case: A reviewer asks "why did the suite stay green while that shipped?" — the Skill scopes to the PR diff, applies mutations to prove red, and returns blocker/major/minor findings with locations. ## Quick Start Ask the agent to review the tests in the current pull request diff and report whether any of them can actually fail.
