Review AI-generated code

To review AI-generated code, Proofline checks it against your production environment, from the terminal or on the pull request. Coding agents write correct-looking code without seeing that environment.

What a coding agent cannot see

An agent works from the repository and the prompt. It does not know the database's settings, the cluster's network policies, or the variables the production service defines.

Proofline checks each change against a recorded snapshot of that environment. It reviews a pull request the same way whether a person or an agent wrote it.

Review local changes before a pull request

Run proofline review in your repository to review uncommitted or staged changes against their base commit. The CLI can also install Git hooks that review each commit and push.

In continuous integration, proofline ci review can also write its results as SARIF, and proofline findings lists the findings for the current commit.

Connect your coding agent

Connect Codex, Cursor, Gemini CLI, or Kiro through the Proofline CLI, which links the agent to Proofline's Model Context Protocol (MCP) server. The agent can then read a review's status and the reasons an assessment is incomplete.

Each finding offers a fix prompt for your agent. The prompt asks the agent to verify the finding at the reviewed commit before it edits anything. See the coding agent setup.

You can leave pull requests from bot accounts out of automatic reviews, except those from the GitHub Apps you allow.

Questions

Can Proofline review code before I open a pull request?

Yes. Run proofline review to review local changes against their base commit. The result appears with your other reviews.

Which coding agents work with Proofline?

Codex, Cursor, Gemini CLI, and Kiro connect through the Proofline CLI, which links them to Proofline's MCP server.

Does Proofline read my coding agent's instruction files?

Yes. By default, Proofline reads AGENTS.md, CLAUDE.md, and Cursor and Copilot rules at the merge base. An administrator can turn this off.

Review your agent's next change.

See how findings stay quiet or see what the reviewer model can reach.

Get Started with GitHub