Building a Local GitHub-style Review Interface for Claude Code Plans

--- title: Building a Local GitHub-style Review Interface for Claude Code Plans published: true description: GitHub PR Review UI for Claude Code Plans tags: #claudecode #ai #programming # Use a ratio of 100:42 for best results. # published_at: 2026-03-03 23:54 +0000 --- Plan mode in Claude Code feels like reviewing a pull request with no comments, no diff, and no history. Claude thinks for a moment (well much longer than a "moment"), then dumps a wall of text into your terminal - It then asks for approval with many variations of "Yes, ...." Then I open the Markdown file, start taking notes while reading through it. After doing this for a while I felt there should be a better way. So I built [IPE](https://github.com/EduardMaghakyan/ipe).  --- ## What is IPE? IPE intercepts Claude Code's `ExitPlanMode` hook and opens a browser tab showing the plan in a GitHub-style code review interface. You can: - **Add inline comments** on any block or text selection — just like leaving a review comment on a PR - **Click any file reference** (e.g. `` `src/index.ts` ``) to pop open a syntax-highlighted side drawer showing the actual file contents - **Compare plan versions** side-by-side when Claude revises after your feedback - **Switch between sessions** if you're running multiple Claude Code instances at once - **Approve or request changes** — clicking "Request Changes" bundles your inline comments and sends them back to Claude --- ## The Problems I wanted to fix **You can't annotate.** If step 3 looks wrong and step 8 is fine but needs a tweak, you're writing one blob of feedback hoping Claude parses it all correctly. **You lose context across revisions.** Claude revises the plan based on your feedback - but there's no diff. Did it actually address your concern? You're re-reading the whole thing from scratch. --- ## How It Works IPE registers itself as a Claude Code hook on `PermissionRequest` with the `ExitPlanMode` matcher. When Claude finishes planning and tries to proceed, the hook fires. The binary spins up a local HTTP server, opens your browser, and **blocks** - Claude is sitting there waiting for your response. You review at your own pace (the timeout is 4 days, so no rush). When you approve or request changes, the server sends the response back to the hook and the browser tab closes automatically. The whole thing is a self-contained binary built with Bun. --- ## Install in One Line **macOS / Linux:** ```bash curl -fsSL https://raw.githubusercontent.com/eduardmaghakyan/ipe/main/install.sh | bash ``` **Windows (PowerShell):** ```powershell irm https://raw.githubusercontent.com/eduardmaghakyan/ipe/main/install.ps1 | iex ``` That's it. The script downloads the binary and registers the hook in your Claude Code settings automatically. Run it again to update. Verify it's wired up by running `/hooks` inside Claude Code — you should see the `ExitPlanMode` hook listed. --- ## The Workflow in Practice 1. Work with Claude Code as normal 2. Claude generates a plan and calls `ExitPlanMode` 3. Your browser opens with the plan displayed 4. Read through it, click file references to inspect code, leave inline comments where needed 5. Hit **Accept** → Claude proceeds 6. Hit **Request Changes** → your comments go back to Claude, it revises, you review the diff --- ## Try It Out The project is open source: [github.com/EduardMaghakyan/ipe](https://github.com/EduardMaghakyan/ipe) If you use Claude Code in plan mode regularly, give it a spin and let me know what you think. Issues and PRs welcome - there's a lot of room to grow this (comment threads, keyboard shortcuts, plan history persistence...).