I Built a Claude Code Skill That Makes AI Agents Prove Their Answers

Описание и применение

Repo X-Ray позволяет AI‑агентам подтверждать ответы, используя доказательства из кода, конфигураций, тестов, документации, истории Git и обсуждений GitHub, уменьшая риск ложных утверждений.

Задача

AI coding agents can find relevant code but struggle to distinguish evidence from interpretation, leading to uncertain or contradictory answers.

Результат

Repo X-Ray enables agents to provide evidence-backed answers, reducing false certainty and allowing engineers to assess contradictions and historical context.

Рабочий процесс

• 1. Define claim and search scope
• 2. Perform deterministic discovery across source code, configuration, tests, documentation, Git history, and GitHub
• 3. Record evidence and provenance in an evidence ledger
• 4. Apply targeted reasoning over collected evidence
• 5. Validate claim against evidence and classify outcome
• 6. Return structured result with classification and evidence references

Доказательства

Repo X-Ray is a Claude Code skill for evidence-grounded repository analysis

Источник

The important part is the ordering. I don't want the model to invent an explanation first and then search for something that supports it.

Источник

Repo X-Ray uses four classifications: VERIFIED, CONTRADICTED, UNVERIFIED, UNCHECKABLE

Источник