What problem does it solve? Coding agents report success even when their changes are broken, because nothing in the repo contradicts them. This Skill audits whether a repository gives an agent a reliable way to verify its own work — a fast, decisive, specific feedback loop — and identifies exactly where that loop is open. ## Core Features & Use Cases - Hands-on loop verification: Make a small change, then deliberately break something to test whether the suite actually catches regressions, rather than judging tests by reading them. - Structured finding taxonomy: Classifies issues as Blocking (no runnable verification, green-when-broken suites, flaky tests, exit-code swallowing), Worth fixing (slow fast path, untestable seams, missing fixtures), or Optional. - Actionable reporting: Every finding includes a file location, what goes unverified, and a concrete fix, sequenced by loop-closing value rather than coverage gain. - Use Case: Before enabling autonomous or unattended agent runs on a repository, audit whether the agent can tell — without a human — whether each change it makes is correct, and fix the gaps first. ## Quick Start Ask the agent to audit this repository's test setup for agent usability and report what an agent cannot verify on its own.
