A new GitHub repository introduces two SKILL.md files that enable AI coding agents to design and execute claim-driven tests for distributed and stateful systems. The skills work with Claude Code, Codex, Copilot CLI, and other agents that process Markdown and shell commands, producing structured test plans and findings reports with nine-state verdicts.

Traditional integration testing often misses critical distributed system bugs like network partitions and crash-recovery issues. These skills enforce a workflow starting from product claims rather than arbitrary test cases, with each scenario designed to falsify specific claims under fault conditions. The approach binds abstract models like registers, queues, and logs to operation-history schemas with explicit checkers.

The system generates comprehensive artifacts including test plans with coverage arguments and findings reports with SUT/harness/checker/environment blame classification. Reviewers can make ship decisions without re-running tests. Every PASS must cite oracle evidence, and FAIL verdicts include reduction plans for root cause analysis.

This addresses a fundamental gap in distributed system verification where ad-hoc testing leaves critical failure modes unvalidated. By making coverage adequacy a deliverable rather than an afterthought, the skills encode decades of distributed systems testing wisdom into reproducible AI workflows.