taovc/pr-cockpit

What it solves

PR Cockpit streamlines the process of reviewing and fixing pull requests (PRs) by using AI to analyze code changes locally. It eliminates the need to review PRs one by one in a terminal and prevents the fragmentation of fixes across multiple shells, while ensuring that code never leaves the user's machine.

How it works

The tool pulls a repository's PR queue via the GitHub CLI (gh) and creates isolated git worktrees for each PR. An AI agent (powered by Claude or Codex) reviews the code in these read-only worktrees, identifying issues and suggesting fixes. Users can then gate these findings—choosing which ones to post as professional English comments on GitHub—and use a persistent chat to have the AI edit the worktree and push fixes directly to the PR.

Who it’s for

It is designed for developers and maintainers who manage multiple pull requests and want to leverage AI for code review and feature development without sacrificing privacy or control over what is posted to GitHub.

Highlights

• Local-First AI Review: Runs on the user's own Claude or Codex subscription locally, ensuring code privacy.
• Isolated Worktrees: Uses git worktrees to keep PR reviews and AI edits separate from the main development branch.
• Human-in-the-Loop: Requires explicit user approval before any AI finding is posted as a GitHub comment or any code is pushed.
• Integrated Fix Chat: A persistent agentic chat that can edit the PR worktree and commit changes with user confirmation.
• Project Assistant: Supports directory-level chat and the creation of new feature branches for development.
• Multi-Provider Support: Configurable per project to use either Claude or Codex models.