CLI

Use Rerove from your terminal. Install the rerove command, sign in, and bind a repository to a project.

The rerove command reads and writes the same decision ledger the dashboard shows. A CI job or a coding agent runs it and reads its JSON.

Install

On macOS or Linux, run one of these:

brew install rerove/tap/rerove
curl -fsSL https://install.rerove.app | sh

Homebrew puts rerove on your PATH. The script does the rest of this section.

The script downloads the binary for your platform, checks its SHA256 against the published SHA256SUMS file, and puts it in ~/.local/bin. If that folder is not on your PATH, the script prints the line to add to your shell profile.

Two variables change what the script does. REROVE_VERSION pins a release and turns off automatic updates. REROVE_INSTALL_DIR changes the folder.

curl -fsSL https://install.rerove.app | REROVE_VERSION=v0.1.0 sh
curl -fsSL https://install.rerove.app | sudo REROVE_INSTALL_DIR=/usr/local/bin sh

Manual download

Each release on the releases page holds one file per platform and a SHA256SUMS file. Download both. Then check the sum, make the file executable, and move it onto your PATH:

shasum -a 256 -c SHA256SUMS --ignore-missing
chmod +x rerove-macos-arm64
mkdir -p ~/.local/bin
mv rerove-macos-arm64 ~/.local/bin/rerove

The script does not support Windows. Download the .exe and put it on your PATH.

Sign in

rerove login

The command opens your browser. Sign in to Rerove there. The command waits, then stores a session token in ~/.config/rerove/. You sign in once. Every later command reuses the token, and so does a coding agent that runs rerove for you.

If a command needs a login and you have none, the CLI asks on a terminal: "Log in now?" Enter starts the browser login. In a pipe, with JSON output, or with CI set, the command exits 3 instead.

rerove status   # which account is signed in
rerove logout   # delete the token

Bind a repository to a project

Run this anywhere inside the checkout:

rerove init

init does six things:

1. Finds the repository on the server by its origin remote.
2. Puts the repository in a project. If the dashboard already set one, the file follows it, and init asks nothing. If not, a picker lists your projects. An admin can create one from the last entry.
3. Asks to turn on documents and reviews. Enter means yes. It asks only about a switch that is off. Reviews spend credits.
4. Writes rerove.json at the repository root with the project_id. An existing file keeps its ignore_patterns, author lists, and trigger_mode. A file the server would reject does not stop init. Each problem becomes a checklist item, and the skill fixes it.
5. Installs the agent skill when the checkout holds a file a coding agent left, such as AGENTS.md, CLAUDE.md, or .cursorrules. When claude or codex is on your PATH, it asks to run the skill now. See Set up the agent skill.
6. On a Homebrew install, asks once whether to update in the background. See Update.

Then it prints what is left: a connected repository with documents and reviews on, a project, and a valid rerove.json on the default branch. Commit the file. The server reads it from the default branch only.

To see the checklist again, run rerove doctor --repo. See Check the setup.

Set up the agent skill

Rerove reads every markdown file on the default branch into the decision ledger. Test fixtures, generated docs, and changelogs add noise. Which file is product knowledge is a judgment call, so a coding agent makes it. The rerove-setup skill tells the agent how.

rerove init installs the skill when it finds an agent file. To install it by hand, or again after an update, run:

rerove skill install

It writes .agents/skills/rerove-setup/SKILL.md, which Codex, Cursor, Gemini CLI, and Copilot read. When the checkout has a .claude/ folder or a CLAUDE.md file, it also writes .claude/skills/rerove-setup/SKILL.md for Claude Code. --claude adds that copy anywhere. A copy you edited is kept. --replace overwrites it.

Then open your coding agent in the repository and run the skill:

The skill reads the checkout only. It needs no network and no CLI. First it fixes a rerove.json the server would reject. Then it lists the markdown files, judges each folder, and asks you when it is not sure. Then it writes ignore_patterns into rerove.json and stages the file. Commit the skill file too, so the whole team gets it.

A run that init starts asks nothing. It keeps every file it is not sure about and lists them at the end. Run the skill in your agent to clear those doubts.

Check the setup

When a command fails, run:

rerove doctor

It prints the auth state, environment, project, CLI version, and update state. The report needs no network. ok is true when the next command should reach the server with no action from you.

To check a repository, run this inside the checkout:

rerove doctor --repo

It prints the same checklist as init, and writes nothing. Each item says what is left. Documents that are off count as an item. While an item is left, the command exits 8, so a CI job fails until the setup is done:

rerove doctor --repo                  # exit 8 fails the job
rerove doctor --repo || [ $? -eq 8 ]  # under set -e, allow exit 8

Below the checklist, it says whether the documents became statements. It shows the last scan, the file count, the extraction state per file, the failed files, and the statement counts. With no accepted statement, it names the reason. In JSON, knowledge.reason holds a code, such as scan_running or statements_pending. knowledge.wait is true when the state clears by itself, so an agent checks again later. knowledge.wait_max_seconds caps that wait.

--repository-id ID names the repository, as in init.

Commands

A command finds its project in this order:

1. The --project-id flag.
2. rerove.json at the checkout root, or the nearest one above the current folder.
3. Without both, an error that says to run rerove init.

A list command takes --page and --page-size, from 1 to 100. The JSON holds total and has_more.

Output and exit codes

On a terminal, rerove prints tables. In a pipe, it prints JSON. --format json or --format table forces one.

The JSON has a stable envelope:

{
  "schema_version": "1",
  "kind": "statements",
  "data": { }
}

schema_version changes only when the envelope shape changes. A script can pin to it. A table can change between releases.

An error in JSON mode goes to stderr in the same shape, with an error object and the exit_code.

Environment variables

Update

The CLI updates itself. Once a day, a command checks github.com for a newer release. The check runs in the background and sends only the CLI version. When a newer release exists, the same background process installs it. The command you ran does not wait. The new version runs from your next command, and on a terminal it prints what changed. To read the notes again, run rerove changelog.

A Homebrew install updates itself only when you allow it. rerove init asks once on a terminal. If you say no, the CLI prints a notice once a day. To change the answer, run:

rerove config set homebrew_auto_update true

To update now:

rerove update

On a Homebrew install, it runs brew upgrade rerove/tap/rerove. Both paths check the download against the SHA256 the release publishes.

If an automatic update fails, for example because the install folder needs root, the daily notice shows the error. rerove doctor shows it in update_error.

To turn automatic updates off on a direct install, run rerove config set auto_update false. REROVE_NO_AUTO_UPDATE=1 turns them off on every install. REROVE_NO_UPDATE_CHECK=1 turns off the check too.

Report a problem

Open an issue at github.com/rerove/cli/issues, or write to support@rerove.app. Add the output of rerove doctor.

See also

• Configuration
• MCP server
• AI agents