What problem does it solve? QA work tied to a ticket or a just-finished code change often ends in unverified claims: tests written from the diff that bind to one implementation, green suites that never proved they would catch a regression, and verdicts posted with no evidence behind them. This Skill turns a ticket, branch, or developer request into verified evidence, durable regression tests, and sentinel tests for every defect found. ## Core Features & Use Cases - Structured 10-step QA sequence: ticket intake, PR review-state check, worktree isolation, diff review, environment access, interaction-driven understanding, durable test authoring, negative control, adversarial review, and reporting. - Negative control enforcement: runs the new suite against an environment without the fix (or the merge-base commit locally) to prove tests actually discriminate the change. - Adversarial review and evidence bundles: dispatches subagent reviewers to attack the tests, scores testing rigour, and produces per-ticket evidence bundles via companion-mode. - Use Case: A developer says "QA this ticket before I open the PR." The Skill derives acceptance criteria from the diff, confirms them with the developer, drives the app in a browser, writes one test per AC plus sentinels for defects, proves the suite fails on the merge-base, and reports a grounded verdict. ## Quick Start Ask the agent to QA a specific ticket or branch, for example: "QA ticket ABC-450 on its feature branch and verify the acceptance criteria with evidence before I open the PR."
