How do you know what your AI coding agent actually did?

Coding agents like Claude Code, Cursor and Codex now run commands, read files and call tools with our own permissions. After a long session, most of us couldn't say exactly what happened: the transcript is a conversation, not a record.

Curious how others handle this:

- Do you review what the agent ran, or just the final diff?

- Has an agent ever done something you only noticed later?

- For teams: is anyone keeping a real audit trail of agent actions, or is it trust plus code review?

I've been working on this problem and launched an open-source take on it today (Agentmetry: a local, tamper-evident log of every tool call). I'd honestly rather hear how you handle it than talk about mine.