How to audit the changes Cursor’s AI makes

By the AgentReceipt team · Updated 13 November 2026

When Cursor’s agent works on a task, the result is spread over many files, and the editor’s own history is not something a teammate can review. Here is a practical way to keep a record you can check and share.

What goes missing by default

Your pull request shows the final diff. It does not show what you asked Cursor to do, which files the agent looked at, or whether it touched code that has nothing to do with the task. Those are exactly the things a reviewer needs.

Set up a record for Cursor

AgentReceipt works with Cursor as well as Claude Code. Install the command-line tool, then initialise it in your repository for Cursor.

npm install -g @agent-receipt/cli
cd your-repo
agentreceipt init --agent cursor --pre-push

Check before you push

The --pre-push option installs a git hook that runs agentreceipt check --hook before each push. If the session broke a rule, such as a secret-like value in the diff, you find out on your own machine instead of in review.

You can also run agentreceipt check yourself at any time. It checks the latest session against the rules and tells you what it found.

agentreceipt check

Use more than one agent in the same repository

Many teams use Cursor for some tasks and Claude Code for others. You can set up several agents in one repository with agentreceipt init --agent claude-code,cursor, and the receipts keep track of which agent made which change.

agentreceipt init --agent claude-code,cursor --pre-push