Per-Agent Review

Review Each Agent's Work Separately

When five agents modify fifty files, reviewing everything at once is overwhelming. AgentsRoom tracks which agent changed which files. Filter by agent, review their work in isolation, and commit each contribution separately.

Screenshots

Real captures of the app, added as this feature evolves.

  • Changes tab: hovering Pull lists the commits waiting on the remote
    See which commits a pull or a push will bring before running it, and open any of them with its diff.
  • Non-code changes section of the Changes tab with its Clear the list confirmation
    Empty the list of non-code changes without a commit: nothing is undone, the version history keeps them.
  • Per-Agent Review page section about non-code changes
    The Per-Agent Review page explains the non-code changes and how to clear their list.
1 / 3

A real session filmed in the app: three AI coding agents, two on Claude Code and one on Codex, each change one file of the same project. The Changes list gives each agent its own tab, and a commit takes only that agent's files: three agents, three commits in the history.

Watch on YouTube

Review what the agent changed

Each agent's diff, file by file, with your comments on a line sent straight back to the agent.

AgentsRoom Changes tab showing the diff of one AI agent, with a comment on a line

How Per-Agent Review Works

Every tool you need to review multi-agent code changes without losing track of who did what.

Agent Filter Tabs

When two or more agents have modified files, tabs appear at the top of the review panel. Each tab shows the agent's avatar, name, and file count. Click a tab to see only that agent's changes.

Automatic File Attribution

AgentsRoom attributes files two ways. A hook fires on every Edit, Write and NotebookEdit call and records the file as the agent writes it. For files written by a shell command, AgentsRoom knows when each agent ran its commands and matches that against the file's modification time.

Scoped Commits Per Agent

When an agent filter is active, the commit action only stages that agent's files. Each agent's work gets its own commit with a clear message. Your git history stays organized by contributor, even when agents work in parallel.

Filtered Keyboard Navigation

Navigate through the filtered file list with arrow keys or j/k shortcuts. The progress bar updates to reflect only the files in the current filter. Mark files as reviewed one by one until the filtered set is complete.

Auto-Follow Active Agent

Switch to an agent's terminal and the file list auto-filters to show that agent's recent changes. No need to manually select the filter. Override at any time by clicking a different tab, or turn auto-follow off in View options to keep the filter and worktree you picked.

Avatars on Shared Files

A file written by one agent is already named by its filter tab. When two or more agents modified the same file, its row shows all their avatars side by side. Hover an avatar to see the agent's name, click it to jump straight to that agent.

Non-Code Changes, Reviewed Too

Agents also change what git never sees: a prompt, a skill, a memory note, a trigger, a ticket. The Non-code changes section under the files lists them, each with a line diff and Restore the previous version. Once you have read them, Clear the list empties the section without a commit, and nothing is undone.

From Parallel Work to Clean Commits

Four steps from multi-agent chaos to an organized git history.

1

Agents work in parallel

Multiple Claude Code agents run simultaneously on your project. Each agent focuses on its assigned task and modifies the files it needs.

2

AgentsRoom tracks every change

As each agent works, AgentsRoom records which files it touches, whether the agent used a file tool or wrote through a shell command.

3

Filter by agent in review mode

Open review mode and click an agent's tab. The file list narrows to show only that agent's modifications. Review their diffs, ask questions about specific changes, and mark files as reviewed.

4

Commit per agent

Once you have reviewed an agent's files, commit only their changes. The commit stages just those files. Repeat for each agent to build a clean, well-attributed git history, then click Create PR: the AI proposes the pull request's title and description from the branch diff, and you edit them before anything is sent.

AgentsRoom per-agent review: filtering code changes by individual AI agent with scoped commit tabs
AgentsRoom Changes tab: a file modified by two AI coding agents shows both agents' avatars
A file changed by two agents shows both avatars. Click one to open that agent.
How does AgentsRoom know which agent modified which file?
AgentsRoom uses two complementary methods. First, a hook on the agent's file tools (Edit, Write, NotebookEdit) records each file the moment it is written. Second, for files written by a shell command, which no file tool ever sees, AgentsRoom reads each agent's session log to know when its commands ran and matches that against the file's modification time. When two agents could both have written the same file, it stays unattributed rather than credited to the wrong one. With Claude Code, an AgentsRoom plugin also compares the repository right before and right after each shell command, which catches deleted files too and never counts a read-only command.
Can I review all changes at once, without filtering?
Yes. The first tab in review mode is always 'All files', which shows every modification across all agents. Agent filtering is optional. You can review everything together and still commit all changes in a single commit.
What happens if two agents modify the same file?
The file appears under both agents' tabs, and its row in the changes panel shows both agents' avatars, so a shared file never slips into the wrong commit unnoticed. Click an avatar to open that agent directly. You then decide which agent's commit includes the file, or commit it with all changes at once.
Does this work with any number of agents?
Yes. Agent filter tabs appear as soon as two or more agents have attributed files. The system works the same with 2 agents or 12. Each agent gets its own tab with an accurate file count.
Can I commit all agents' changes together instead of separately?
Yes. Select the 'All files' tab and commit from there. All modified files are staged together in a single commit. Per-agent commits are an option, not a requirement.
Can I throw away a change an agent made, without typing a git command?
Yes, since version 1.203.0. In the Changes tab, the "..." menu of a modified file has Discard changes. A first click says what will be lost, a second click runs it: a tracked file goes back to its last committed version, index and working tree included, and a file that was never committed is deleted from disk. Nothing is recoverable afterwards, which is why it takes two clicks. The same action exists in the mobile app, with a long press on a file of the Changes screen. It acts on the repository, whatever CLI produced the change.
Can I create the pull request from AgentsRoom, follow its CI and review, and hand the failures and comments to an agent?
Yes. Next to Push, the Changes tab shows the GitHub Actions or GitLab CI status of the pushed commit (running, passed, failed), read with the account you connected for Import repositories. When it failed, the badge lists the failed jobs with the end of their log, and Send to agent to fix gives them to an agent already working in that checkout, or starts a new one. Beside it, Create PR (Push and create PR if the branch is not pushed yet) opens a pull request on GitHub or a merge request on GitLab, gitlab.com or self-hosted: the AI proposes the title and the description from the branch diff, its commits and the linked ticket, and you edit them, change the base branch or tick Open as a draft before anything is sent. The badge then shows the state (draft, open, merged, closed), the review verdict and the number of unresolved comments, and Send the comments to the agent hands those comments, with their file, line, author and code excerpt, to the agent of that checkout in one message: it fixes them and pushes to the same branch, without merging. Badges and actions are also in the mobile app, and AgentsRoom can notify you, on the desktop and on your phone, when a pipeline finishes or a review arrives on your pull request (Settings, Notifications). Merging, approving and resolving a thread stay on GitHub or GitLab.
Can I clear the non-code changes without committing?
Yes. In the Changes tab, the Non-code changes section lists what agents changed outside the code: prompts, skills, memory notes, triggers, tickets and their other AgentsRoom actions. A commit takes them out of the list. To empty it without a commit, click the Clear the list icon in the section header, then Clear. The changes leave the section and the tab's count, but nothing is undone or deleted: they stay applied, and the version history keeps them. On an agent's tab, only that agent's changes are cleared. Clearing cannot be reverted and restores nothing: to undo one change, open it and click Restore the previous version. This section is in the desktop app only.
Can I see which commits a pull or a push will bring before running it?
Yes. The Changes tab runs Fetch, Pull, Merge and Push from the buttons next to the branch name, with no agent and no terminal. After a Fetch, Pull shows how many commits are waiting on the remote, and Push how many of yours are not pushed yet. Hover either button and a card lists those commits, newest first, with their author, subject, short hash and age: exactly the commits the button counts. Show more loads the rest. Click a commit to open it in the commit window, with its changed files and its diff, the same window as the history. This card is in the desktop app only, since a phone has no hover.

Further reading

Review Multi-Agent Code With Clarity

Download AgentsRoom and see exactly what each AI agent changed in your codebase. Free for up to 2 projects.

Download AgentsRoomFree

Companion app: monitor your agents on the go

Bring your own: Claude, Codex, Antigravity CLI, or other AI provider.

Push bugs and requests straight to your public backlog.