Getting started

ReviewFixLoop runs several coding agents as reviewers of one pull request, while your own agent fixes what holds. Reviews run on your machine, with agents you installed and signed in to yourself.

What you need

• macOS or Linux, on arm64 or x64.
• git, and the GitHub CLI signed in (gh auth status succeeds).
• At least one supported agent, installed and signed in: Claude Code, Codex, GitHub Copilot CLI, OpenCode or Antigravity. Reviews use your own plans and API keys with those vendors; see vendor terms.
• A GitHub account at least 30 days old. You must be 18 or over, and the Service is not offered in the European Economic Area, the United Kingdom or Switzerland; see the terms.

Install

bash -o pipefail -c 'curl -fsSL https://reviewfixloop.com/install.sh | sh'

If ~/.local/bin was not on your PATH, the script adds it to your shell profile. Open a new terminal, or run export PATH="$HOME/.local/bin:$PATH", and then:

review-fix-loop login
review-fix-loop install

• The script puts the executable at ~/.local/bin/review-fix-loop, with the alias rfl unless another rfl is on your PATH.
• login shows a code and opens a browser page to approve it. In a terminal, it also asks which of the agents on this machine are your default reviewers.
• install adds the review-fix-loop skill to the agents on this machine. For Claude Code it also offers to allow the CLI to run without asking each time.

Review a pull request

Push your branch, open a pull request, and invoke the review-fix-loop skill in your agent (/review-fix-loop in Claude Code). Your agent starts the run, waits for findings, fixes the ones that hold, judges each one, and stops when no reviewer has anything left to find. Follow the run on the page the agent links to.

If you would rather not review inside your running agent, start a run with a separate agent as the resolver:

review-fix-loop review --resolver claude-code --purpose 'Why this pull request is wanted'

The first run of a pull request needs --purpose or --requirements-file; later runs of the same pull request keep that record.

Choose reviewers

Reviewers are <harness>[:model][@effort], for example codex:gpt-6.1-sol@high. A run uses the first of these that names any:

1. --reviewers on review.
2. reviewers in the repository's .review-fix-loop.jsonc.
3. reviewers in your ~/.config/review-fix-loop/config.json.
4. Your workspace's default reviewers (review-fix-loop workspace reviewers --workspace <slug>).

At most once a month, the workspace's page may recommend to its owner one change to the default reviewers: replacing one with a reviewer that, in the runs both took part in, found more major defects the other default reviewers missed, or removing one that found no such defect. It shows the counts behind the recommendation, and the default reviewers change only when the owner approves it.

Configure a repository

Commit .review-fix-loop.jsonc at the repository's root. Unknown keys are refused. trust, sessionTimeoutMinutes, sandbox and prSummary are read from the pull request's base branch, so they take effect only after they are merged there.

{
  "reviewers": ["claude-code", "codex:gpt-6.1-sol@high"],
  // Finding tags not to report.
  "excludedTags": ["doc"],
  // Whose pull requests to review: everyone, collaborators, or self.
  "trust": "collaborators",
  // The time limit of one review pass, in minutes.
  "sessionTimeoutMinutes": 60,
  // Post each run's summary as a comment on the pull request.
  "prSummary": true,
}

trust protects you from code in pull requests: reviewers run that code with your authority. It defaults to collaborators on public repositories and everyone on private ones.

"prSummary": true makes the CLI comment on the pull request with your own gh when a run ends: what was fixed, what was rejected and why, what remains unresolved, and where a human reviewer should look. It updates its earlier comment instead of adding another. Anyone who can read the pull request can read the comment, everyone on a public repository, so it is off by default.

"sandbox": true runs reviewers away from the machine's other files and unlisted hosts, at some cost to review quality. It needs more than a normal sign-in: of the agents above only Claude Code and Codex run in it, and each needs a sign-in environment variable (review-fix-loop manifest show <harness> lists it). See what the sandbox needs, including on Linux. review --resolver is refused.

Who can see your runs

Only the members of a run's workspace can read its findings, on public repositories too. A run can count in the aggregates of the leaderboard. See where your data goes before your first run.

A workspace's owner can have the service keep each run's session logs, secrets in common formats masked, for the workspace's members to read on the run's page: review-fix-loop workspace session-logs set on --workspace <slug>, or the workspace's page. It is off by default.

Commands you may need

For anything else, write to contact@willbooster.com.