By the time a manual code review catches a hardcoded API key or a retry loop with no exit condition, an AI coding agent has already written it to disk — and possibly already run it. Two freshly shipped open-source tools — Agent Verifier and Conduct — close this window by adding automated pre-execution checks that run before your agent touches anything: before files are written, before commands execute, before the damage is done. This tutorial walks through setting up both tools, the four error classes they catch, and how to combine them into a two-stage review layer alongside Claude Code or any other coding agent. TL;DR : Agent Verifier runs static checks on your agent's pending actions and flags hardcoded secrets, unbounded loops, hallucinated tool references, and context-blowing prompts before a session runs. Conduct intercepts each action in real time with a separate reviewer agent that evaluates session context, the pending action, and the current file state before passing or blocking.…