AgentReceipt

See exactly what Claude Code changed in your repo

By the AgentReceipt team · Updated 13 November 2026

Claude Code can edit many files in one session. Before you merge, you want to know what it changed, why it changed it, and whether anything in there is risky. Here is how to answer that, from the simplest method to the most complete.

Start with git, and know its limits

git status and git diff show the end result: which files differ and how. That is the right first look. What git cannot tell you is the prompt that caused the change, the commands the agent ran along the way, or which edits belong to which request.

If the agent made several commits, git log -p helps, but commit messages written by an agent often describe the intent rather than check it.

The session transcript shows the why, but it is not attached to your code

Claude Code stores session transcripts on your computer. They show what you asked and what the agent answered. The weakness is that a transcript is long, it lives outside the repository, and a reviewer looking at your pull request never sees it.

Hooks record it as it happens

Claude Code can run a command when events happen in a session, such as a prompt being submitted or a file being edited. A tool that listens to those events can write down the prompt, the files and the diff while they are fresh, instead of reconstructing them later.

AgentReceipt installs those hooks for you. It merges them into .claude/settings.json and creates a .agentreceipt/ folder in the repository.

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

What a receipt contains

When a session ends, AgentReceipt builds one receipt for it. It holds:

  • the prompt, with secret-like values removed before anything is written to disk
  • every file changed, with lines added and removed
  • flags, each with the reason it was raised, from fixed rules rather than a model’s opinion

Read the receipt, not the whole transcript

Open the latest receipt for the current commit with agentreceipt show. A flag such as a weakened test or a new dependency tells you where to look first, so a long session becomes a short review.

agentreceipt show