Show HN: My coding agent apologized for being right, so I built Frank (github.com)

A new local plugin for AI coding agents called "Frank" has been announced, designed to enhance the interaction between developers and AI by encouraging evidence-based coding responses. Frank implements two primary behaviors: it ensures that when an answer is disputed, the agent provides concrete evidence to support its verdict instead of giving an apologetic response, and it refrains from claiming the task is complete without a verification receipt or a clear acknowledgment of unverified status. The plugin operates on popular coding platforms like Claude Code and Codex, utilizing local "Stop hooks" to cross-check final messages against a ledger of verification commands. The significance of Frank lies in its potential to improve coding practices by reducing erroneous confidence in AI-generated code. Previous benchmarks showed that conventional AI agents frequently apologized for mistakes, leading to misunderstandings rather than clarity. Frank demonstrated a significant improvement, with results indicating that it maintained accuracy in responses while effectively managing disputes and relying on factual validation before claiming task completion. This innovative approach not only enhances accountability in AI tools but also makes them more robust by relying on a system of validation rather than unverified assertions.