How to Audit What Your AI Agent Actually Did After the Session

When you hand off a multi-hour task to an AI coding agent and come back to the results, the right question isn't "did it finish?" — it's "did it stay within scope?" Agents running Claude Code, Codex,