What problem does it solve? It prevents unverified claims like "tests pass" or "this fixes it" by enforcing a strict evidence discipline: predictions written before running commands, fresh command output compared against predictions, and honest labeling of anything unobserved. ## Core Features & Use Cases - Evidence Bar: A four-part test that separates real evidence (predicted, run, compared, pasteable output) from reasoning-as-evidence or pasted expectations. - Three Gates: Documents the repo's lint, vitest, and Next.js build gates with exact predicted outputs and known baselines. - Test-Writing Runbook: Step-by-step instructions for adding vitest files under lib/tests, including env snapshot/restore patterns and placement verification with npx vitest list. - Use Case: Before writing "this fixes the share TTL bug" in a PR description, use this Skill to write the done-criteria, add a regression test for clampShareHours, run the gates, and paste the actual output as evidence. ## Quick Start Ask the AI to define done-criteria and the exact verification commands with predicted outputs for the change you are about to make.