Use cases
Six reviewers, one PR.
Point the fleet at a large branch and give each session a different angle: correctness, security, tests, performance, API surface, docs. Reviewers that do not share blind spots.
Divide the angles
One concern per session.
Specialists, not a chorus
The Conductor splits the review into a task per angle: one session traces logic and edge cases, one probes input boundaries, one checks what the tests actually cover, one reads the diff purely for the API it leaves behind.
Reviewers that run code
Each reviewer is a real Claude Code CLI session in your workspace with a native PTY. It can run the test suite, execute the changed paths, and check claims against behavior instead of only reading the diff.
Checked before it counts
Loom verifies work before a task is marked done, so a session that skimmed the branch and declared it clean gets re-checked rather than waved through.
The brief
Ask for findings you can act on.
Have each session write its findings to a file with paths and line references. Loom's markdown preview renders the reports in place, and the built-in editor opens any file a finding points at.
# one branch, six angles
Review the branch feature/billing-rework.
One session per angle: correctness, security,
test coverage, performance, API surface, docs.
Write findings to review/<angle>.md with file
and line references. Run the tests, do not
assume they pass.Honest scope
A first pass, not a rubber stamp.
Does this replace human review?
No. It is a thorough first pass that surfaces issues across angles a single reader would have to hold at once. You read the findings, you read the branch, and you stay the approver.
Where do I read the branch myself?
The built-in source control panel shows diffs hunk by hunk with a full git graph of the history, so you can walk the branch commit by commit without leaving Loom. See source control.
Can the reviewers change the code?
That is your call in the brief. Ask for findings only, or ask sessions to also fix what they flag. Every file, git, and shell operation runs inside the authorized workspace registry, and auto-accept is toggleable per terminal. See permission modes.
Hand it the work.
Walk away.
macOS, Linux, and Windows. Around 13 MB. Free and open source.