Проверка результата перед отчётом об успехе

engineering

verification-before-completion is a Claude Code skill that requires the agent to run verification commands and confirm their actual output before making any claim that a task is done, a bug is fixed, or tests are passing. The skill activates at the moment an agent would otherwise declare success — instead, it enforces an evidence-before-assertions rule, blocking optimistic statements until real results are observed. This prevents the common AI pattern of reassuring developers without checking the actual state of the code. It fits naturally into pre-commit and pre-PR workflows, and is especially valuable for senior engineers who want verifiable proof of correctness rather than confident-sounding summaries from an agent.