Claude Code
Review what Claude Code changed.
Claude Code runs in your terminal and works for a long time without you. That is the point of it, and it is also why the pull request it produces is the hardest kind to review: you have no memory of it being written.
The change arrives whole
A terminal agent working autonomously will touch whatever the task requires and commit as it goes. By the time you look, there is a finished branch and a session log, and the reasoning that produced each decision is in the log rather than in the diff. Nothing about the change is annotated with why, so the review has to reconstruct intent from structure.
What to check first
Count the distinct structural changes.
A long run tends to solve the task and also fix two things it noticed on the way. Each is defensible; together they are a change that does four things and cannot be described in a sentence. More than two or three and the honest answer is that it should have been split.
Read the new files for placement, not correctness.
Where does each one belong, what responsibility does it introduce, who calls it, and what does it call? A new file that reaches across a domain boundary is a structural decision made silently, and it is far cheaper to reverse now.
Look for an abstraction that already existed.
The agent has no memory of your codebase's history. A new TokenService in a repository that already has a TokenManager is the most common failure of the category, and it is invisible in the diff because the existing one does not appear in it.
Questions
Does Cutplane integrate with Claude Code?
It does not need to. Cutplane reads the repository and the two commits, so a branch Claude Code produced is analysed exactly like any other. There is nothing to install into your agent and no configuration that names it.
Can I run it on a branch that was never opened as a pull request?
Yes. Give it any two refs, or none — with no base it uses the merge-base with your default branch and the report says that is what it did. Committing straight to a branch without opening a PR is the normal terminal-agent workflow and it is supported directly.
What does it do about the session log?
Nothing. It reads the code, not the transcript. What the agent said it was doing and what the commits actually did are two different artefacts, and the second is the one that ships.
The method behind all of this is in how to review a pull request your AI agent wrote, and a real report on a public pull request is the output on something you can go and check.