Show HN: Makoto the Claude Code plugin that keeps the AI honest (github.com)

Makoto, a new plugin for Claude Code, introduces a robust integrity system that ensures the AI’s claims about its actions are verifiable and accurate. It monitors tool calls and blocks any attempts by the AI to falsely assert actions like running tests or citing papers. Instead of relying on external truths, Makoto assesses the AI based solely on its stated actions and maintains a record against which these assertions are held. If an AI action is claimed but not executed, or if verification processes are bypassed, Makoto intervenes by blocking the action and prompting the AI to correct its statements. This development is significant for the AI/ML community as it addresses concerns around the reliability and accountability of AI systems. By employing detailed pre-checks and end-of-turn gates, Makoto ensures that every claim made by the AI is backed by actual execution. The plugin includes numerous checks (22 distributed over five families) designed to detect discrepancies such as fabricated evidence or verification suppression. This innovative approach not only enhances the transparency of AI operations but also sets a new standard for integrity within automated systems, encouraging further development of ethical AI practices.