Inter-office memo
- To
- Everyone checking what a coding agent changed
- From
- Mo, Chief Executive
- Date
- November 14, 2026, 1:44 AM CST
- Re
- Reviewing Claude Code and Codex changes
How to review Claude Code and Codex changes in 2026: 8 ways compared
Start with your agent’s own review, in its plan: /diff and /code-review in Claude Code, or the ChatGPT app’s review pane and /review in Codex. Hunk and Plannotator are free for both, Warp sends review comments to either, a pull request adds an AI reviewer, and Enjoy shows each reply’s changes on your phone and hands them to another lab’s agent.
Disclosure: Enjoy is our product, and it runs Claude Code and Codex. We checked every way below against its maker’s own pages on November 14, 2026, and we say where another way is the better pick, including the free ones that don’t need Enjoy.
The ways to review an agent’s changes, compared
If you use one lab’s agent, start with its own review, which comes with your plan. In Claude Code, /diff shows what changed and /code-review looks for bugs in it; in Codex, the ChatGPT app’s review pane shows the diff and takes comments on lines, and /review reports what it finds. The other ways are for one diff view across agents, for sending comments back from a browser or terminal, for a second reviewer before anything merges, or for reviewing away from your desk.
| Way | Best for | Price, checked November 14, 2026 | Where you review | Agents |
|---|---|---|---|---|
| Git in a terminal | Seeing exactly what’s on disk, free | Free | Your terminal | Any |
| Claude Code’s own review | Claude Code, in a terminal, the desktop app or VS Code | Included in Claude Pro ($20 a month), Max ($100), Ultra ($200), Team and Enterprise | A panel in the terminal; the desktop app’s diff view, with line comments; VS Code | Claude Code |
| Codex’s review pane and /review | Codex, with line comments and hunk staging | Codex comes with every ChatGPT plan, from Free; the CLI from Plus ($20 a month) | The ChatGPT desktop app; /diff in the CLI |
Codex |
| A pull request on GitHub | Teams, and a second reviewer before merging | Codex reviews from ChatGPT Plus ($20); Copilot code review from Copilot Pro ($10), or free for students; Claude’s Code Review for Team and Enterprise, on usage credits | GitHub | Any agent whose work you push |
| Hunk | A better terminal diff, with the agent’s notes beside the code | Free | Your terminal on Mac, Linux, Windows | Any agent that can use its skill |
| Plannotator | Commenting in a browser and sending it all back | Free | Your browser, served from your computer | Claude Code, Codex and eight more |
| Warp | Reviewing in the terminal your agents run in | Free; Build $20 a month; Max $200; Business $50 a user | Warp, on Mac, Windows, Linux | Claude Code, Codex, Gemini CLI, OpenCode, Warp’s agent |
| Enjoy | Each reply’s changes on your phone, and a second lab’s review | A monthly subscription, free for 7 days, with your phone and one computer (pricing) | Desktop, browser and iPhone | Claude Code, Codex, Grok Build, Antigravity, Muse Code, OpenCode |
Prices are per month as each maker lists them, before tax. Every way here reviews work your agents do on your own AI plans, which their companies bill for separately.
Why reviewing got harder
Three things make it easy to miss what an agent did:
- It stopped asking. Claude Code now starts terminal and VS Code sessions in auto mode, “with Claude Code v2.1.283 or later,” says its permission modes page, and in that mode “Claude edits most files in your workspace without asking,” per its VS Code docs. You no longer see each diff as it happens unless you switch back, as our memo explains, so review moves to the end.
- Undo isn’t complete. Claude Code’s checkpoints restore its own edits, but “A file that only Bash commands changed isn’t tracked and stays as it is,” says its checkpointing page, which adds that checkpoints are “Not a replacement for version control.”
- New files hide. Plain
git diffshows “the changes you made relative to the index,” so a file the agent created, which Git doesn’t track yet, isn’t in it. Hunk’s README puts it plainly for Git’s own diff: “untracked files will not appear there.”
Git in a terminal
Best for: seeing exactly what’s on disk, with any agent, for free.
git status “Displays paths that have differences between the index file and the current HEAD commit, paths that have differences between the working tree and the index file, and paths in the working tree that are not tracked by Git,” so it’s where new files show up. git diff then shows the changes to the files Git already tracks, per its docs. To take some of an agent’s work and not the rest, git add -p lets you “Interactively choose hunks of patch between the index and the work tree and add them to the index,” which “gives the user a chance to review the difference before adding modified contents to the index.” Commit before you hand an agent a task, and everything it does is one diff away.
Price: free.
Watch for: it’s bare. New files need git status, there’s no way to leave a comment on a line for the agent, and you describe any fix in your next message yourself.
Claude Code’s own review
Best for: Claude Code users, wherever they run it.
Claude Code has a review for each place it runs, all documented in its commands reference:
- In a terminal,
/diffwill “Review the changes in your working tree, including the edits Claude has made so far.” In fullscreen rendering, it opens a panel beside the conversation that “stays open and updates while you keep working.” A picker shows “each turn’s edits,” labeledT1,T2and so on, and a file’saskbutton “attaches that file’s diff to your next prompt.” - In the desktop app, the diff view “lets you review modifications file by file before creating a pull request.” “To comment on specific lines, hover over a line and click the +,” then submit them all with Cmd+Enter or Ctrl+Enter: “Claude reads your comments and makes the requested changes, which appear as a new diff you can review.”
- In VS Code, in Manual mode, “it shows a side-by-side comparison of the original and proposed changes, then asks for permission,” with Accept this change and Reject this change under each one.
- A review by Claude.
/code-reviewreviews “the changes on your branch, looks for correctness bugs, and reports what it found,” and--fixhas Claude apply what it found. In the desktop app, Walk through in diff steps through the findings at their lines. - A way back.
/rewind, or its alias/undo, opens a menu of your prompts, where Restore code will “revert file changes while keeping the conversation.”
Price: included with Claude Code. Claude’s pricing has Pro at $20 a month, or $17 a month billed yearly, Max at $100 and Ultra at $200, and Team and Enterprise include it too; the Free plan doesn’t include Claude Code. /code-review “works on any plan” and “counts toward your plan’s usage.” Its deeper cloud review, /code-review ultra, “Includes 3 free runs on Pro, Max, and Ultra, then requires usage credits.”
Watch for: the panel needs “A terminal at least 110 columns wide” and Claude Code v2.1.287 or later, and without fullscreen rendering you get a dialog instead. To approve each edit before it lands, leave auto mode: in a terminal, “From auto, the first press” of Shift+Tab “switches to default,” which Claude Code calls Manual. And /rewind doesn’t catch everything: besides files only shell commands touched, a subagent’s “edits to other files stay, so use version control, such as Git, to revert them.”
Codex’s review pane and /review
Best for: Codex users, especially in the ChatGPT desktop app, where you can comment on lines and keep only part of the work.
Codex’s review pane “reflects the state of your Git repository, not just what Codex edited.” It shows Unstaged changes by default, or “Staged for the Git index, Commit for a selected commit,” and “Branch for the diff against your base branch, or Last turn for the most recent assistant turn.” From there:
- Comment on lines. “Hover over the line you want to comment on,” select the +, and write. “Codex treats inline comments as review guidance,” so “After leaving comments, send a follow-up message that makes your intent explicit.”
- Keep part of it. “Stage, unstage, or revert a single hunk,” or use Stage all and Revert all.
- Ask for a review. “To ask Codex to review the changes, type
/reviewin the composer,” and “Codex reports prioritized findings without changing your working tree.”
In the CLI, /diff will “Show the Git diff, including files Git isn’t tracking yet,” and /review will “Ask Codex to review your working tree.” For scripts, codex review --uncommitted will “Review staged, unstaged, and untracked changes,” per the developer commands.
Price: Codex’s pricing puts it in every ChatGPT plan: Free, Go at $8 a month, Plus at $20 and Pro from $100. Plus adds “Codex on the web, in the CLI, in the IDE extension, and on iOS.”
Watch for: “The review pane requires a project inside a Git repository.” Free and Go get “GPT-6 Luna at Standard speed in the desktop app, subject to rollout,” and the CLI’s /review “uses the current session model unless you set review_model in config.toml,” so a review is only as sharp as the model doing it. The Linux app “is available in preview.”
A pull request on GitHub
Best for: teams, and a second reviewer, human or AI, before anything merges.
Have the agent push its branch and open a pull request, and the people on your team review it the way they review anyone’s. Three AI reviewers can join them:
- Codex. “In a pull request comment, mention
@codex review,” or turn on automatic reviews, says OpenAI’s GitHub page. “In GitHub, Codex flags only P0 and P1 issues so review comments stay focused on high-priority risks.” - Claude. Anthropic’s Code Review runs several agents on each pull request and posts “inline comments on the lines concerned,” for “Team and Enterprise” plans. On any plan,
/code-review --comment“posts the findings on a GitHub pull request as inline comments.” - GitHub Copilot. “Copilot reviews your pull requests, identifies issues, and suggests fixes you can apply in a couple of clicks,” says GitHub’s code review page.
Price: ChatGPT Plus, at $20 a month, includes “Cloud-based integrations like automatic code review.” Claude’s Code Review is “billed on token usage through usage credits, separately from your plan’s included usage,” with “no flat per-review price.” Copilot code review comes with Copilot Pro, at $10 a month, and the plans above it, and with the free Student plan, and “Each time Copilot reviews a pull request or reviews code in your IDE, the interaction consumes AI credits”: GitHub puts an average review at about $1.50 at its default Balanced level, and says “These figures are estimates only.”
Watch for: it reviews work that’s already committed and pushed, so it comes after the fact. AI reviewers miss things: “Copilot is not guaranteed to spot all problems or issues in a pull request,” and Anthropic says “A review with no findings doesn’t mean your code is error-free or secure.” Copilot Free “does not include Copilot code review.”
Hunk
Best for: a better diff in the terminal, where an agent can explain its changes beside the code.
Hunk “is an interactive terminal diff viewer for reviewing code changes,” free under the MIT license, with about 9,800 stars on GitHub; version 0.25.0 came out on November 11, 2026. hunk diff will “review current repo changes, including untracked files,” hunk diff --watch will “auto-reload as the working tree changes,” and hunk gh pr opens a pull request. To bring in your agent, install its skill with npx skills add modem-dev/hunk -g --skill hunk-review, open Hunk, and “Ask your agent to use the hunk-review skill against the live Hunk session.” Its agent docs say the agent can “inspect the same review, navigate it, and leave inline notes,” and “To hand questions back to the agent, leave your own notes with c.”
Price: free.
Watch for: the agent’s notes “start hidden,” so press a to see them. “prebuilt binaries are not published for Windows on arm64 or musl-based Linux such as Alpine,” and for pull requests, “GitHub Enterprise is not currently supported.”
Plannotator
Best for: commenting on a diff in your browser and sending every comment back to the agent at once.
Plannotator is “a local, browser-based review surface for AI coding agents: Claude Code, Codex, Copilot CLI, Gemini CLI, OpenCode, Kiro, Droid, Amp, DeepSeek Harness, and Pi.” Its code review will “Review local changes or remote PRs. Comment on diffs, suggest code. Your comments go back to the agent.” Run /plannotator-review in Claude Code for your uncommitted changes, or give it a pull request’s URL; in Codex, it’s the $plannotator-review skill. It installs on Mac and Linux with a script, and on Windows with PowerShell. It’s dual-licensed under Apache 2.0 or MIT, with about 10,600 stars, and version 0.28.64 came out on November 13, 2026.
Price: free and open source.
Watch for: it checks GitHub for updates when its pages load, and “There is currently no opt-out setting for the release check,” though it sends no review content. Its remote mode “listens on every interface (0.0.0.0) with no login,” so keep that port off networks you don’t trust. On native Windows, Codex’s hooks “are experimental.”
Warp
Best for: reviewing in the terminal your agents already run in, with comments sent straight to them.
Warp’s code review panel “shows all uncommitted changes on your current branch,” or your work against main or any other branch. You can “click and edit the diffs directly,” revert a hunk, and send comments “directly to a running coding agent session, including third-party CLI agents like Claude Code, Codex, and others.” Its review guide says Warp “sends the complete batch at once,” and the agent “applies the requested changes in one pass and returns an updated diff.” Warp is “Available for macOS, Windows, and Linux.”
Price: Warp’s pricing has a Free plan at $0, Build at $20 a month, or $18 billed yearly, Max at $200, and Business at $50 a user a month.
Watch for: the pricing page doesn’t say which plans include the review panel. “Code Review in Warp works with any Git repository,” so it needs one, and Discard all puts files back the way the base branch has them, so Warp advises “make sure you’ve saved or backed up anything you want to keep.”
Enjoy
Best for: reviewing each reply’s changes from your computer, a browser or your phone, and handing them to another lab’s agent for a second look.
Enjoy runs Claude Code, Codex, Grok Build, Antigravity, Muse Code, and local models through OpenCode, on Mac, Windows and Linux, on the plans you already have. For review, it adds:
- Each reply’s changes. When an agent commits during a reply, View changes “shows every file it changed, line by line, with what was added and removed,” and Before / after shows the two versions side by side. “In the iPhone app, each commit appears below the reply with its short code and message.”
- What it did, too. The agent activity panel “lists the commands the agent ran, the files it read and edited, and what happened,” per Enjoy’s guide to checking the work.
- A second lab’s review. Switch the conversation to another agent, and “The new agent gets the conversation so far, including replies from other agents,” so Codex can review what Claude Code just did, or the reverse. For anything important, the guide suggests you “Start a new conversation and ask another agent to review the work.”
- Try the result. Agents “can open your site in Enjoy’s browser beside the conversation, then read the page, click, fill in forms, scroll, and move between pages.”
Price: Enjoy is one monthly subscription, and the first seven days are free, with a card to start. It includes your phone and any browser and one computer; each other computer is $9.99 a month, and each teammate you invite needs a subscription of their own. See pricing.
Watch for: it shows commits, not edits in progress: “Replies without commits don’t have View changes,” so ask your agents to commit, or use one of the tools above for uncommitted work. The Changes window has no comments on lines and no staging or reverting of hunks, and “Enjoy doesn’t have a button that undoes an agent’s changes to your project files.” Your phone connects to the computer doing the work, which needs to be on, online, and running Enjoy. Enjoy isn’t open source, and it’s a monthly subscription on top of your AI plans once the free week ends.
When to stay with your agent’s own review
If you use one agent, its own review is free with your plan and the most complete. Claude Code’s /diff panel and desktop diff view, and Codex’s review pane, with its comments on lines and hunk staging, do more than most tools here. Whatever else you use, keep Git underneath: commit before a task, and every change is reviewable and reversible.
Look further when you want a diff that works the same for every agent (Hunk, Plannotator or Warp), a gate before anything merges (a pull request), or a review from your phone and a second lab’s opinion in the same conversation (Enjoy).
How we compared
We read Anthropic’s Claude Code docs and pricing, OpenAI’s Codex docs and pricing, GitHub’s Copilot docs, Git’s own documentation, Warp’s docs and pricing, and the GitHub pages and docs of Hunk, Plannotator and OpenAI’s Codex plugin for Claude Code on November 14, 2026, and every one is linked below. We quote each maker’s own words for its limits and leave out what it hasn’t documented. These tools change weekly, so we recheck this guide against its sources at least once a month, and the date at the foot of the page says when we last did.
13.Questions.
How do I see what Claude Code changed?
- Run /diff. It shows “the changes in your working tree, including the edits Claude has made so far,” and in fullscreen rendering it stays open beside the conversation, with a view of each turn’s edits. In Claude’s desktop app, the diff view lets you comment on lines, and /rewind restores the code to an earlier prompt.
Does Claude Code show a diff before it edits a file?
- Not by default. Claude Code 2.1.283 and later starts terminal and VS Code sessions in auto mode, where it edits most files without asking. In Manual mode, VS Code shows “a side-by-side comparison of the original and proposed changes” before each edit, and the desktop app shows “a diff and can accept or reject each change.” In a terminal, Shift+Tab once switches from auto to Manual.
How do I review Codex’s changes?
- In the ChatGPT desktop app, open the review pane. It shows unstaged changes by default, or Staged, Commit, Branch or Last turn, and you can comment on lines and stage or revert each hunk. Type /review for “prioritized findings without changing your working tree.” In the CLI, /diff shows the diff, new files included.
Can I comment on a diff and have the agent fix it?
- Yes. Claude’s desktop app, Codex’s review pane, Plannotator and Warp all take comments on lines and send them to the agent, and Hunk lets you leave notes for an agent reading the same review. In plain Git or Enjoy, you describe the fix in your next message.
How do I get Codex to review Claude Code’s work?
- Three ways. OpenAI’s Codex plugin for Claude Code adds /codex:review, “a normal read-only Codex review.” On GitHub, push the branch and comment @codex review on the pull request. In Enjoy, switch the conversation to Codex, which gets the conversation so far, and ask it to review.
How do I undo what an agent changed?
- In Claude Code, /rewind restores code to an earlier prompt, though files only shell commands changed stay as they are. Codex’s review pane has Revert all and per-hunk revert. Either way, Git is the record. Commit before you hand an agent a task, and Git can take you back.