---

name: rr description: Use the rr CLI to sync code and run commands on remote machines. Invoke when the user wants to run tests, builds, or commands remotely, sync files to a remote host, set up remote development, or troubleshoot rr configuration. user-invocable: true allowed-tools:

• Bash
• Read
• Edit
• Grep
• Glob

---

rr (Road Runner) CLI

rr syncs code to remote machines and runs commands there. It handles host failover, file sync with rsync, distributed locking, and test output formatting.

Quick Reference

Copyrr run "make test"     # Sync files + run command
rr exec "git status"   # Run command without syncing
rr sync                # Just sync files
rr doctor              # Diagnose issues
rr monitor             # TUI dashboard for host metrics

Two-Config System

rr uses two config files with different purposes:

Global Config (~/.rr/config.yaml)

Personal host definitions. Not shared with team. Contains SSH connections, directories, and machine-specific settings.

Copyversion: 1

hosts:
  mini:
    ssh:
      - mac-mini.local      # LAN hostname - try first
      - mac-mini-tailscale  # SSH config alias - fallback
    dir: ${HOME}/projects/${PROJECT}
    tags:
      - fast
    env:
      DEBUG: "1"

  server:
    ssh:
      - dev-server          # SSH config alias from ~/.ssh/config
    dir: /var/projects/${PROJECT}

defaults:
  local_fallback: false
  probe_timeout: 2s

Host Options

Reducing Task Verbosity with setup_commands

If you find yourself repeating the same setup in every task (sourcing environments, setting PATH, etc.), move it to setup_commands in the host config:

Copy# ~/.rr/config.yaml
hosts:
  dev-box:
    ssh: [dev.local, dev-tailscale]
    dir: ~/projects/${PROJECT}
    setup_commands:
      - source ~/.local/bin/env     # Load uv, pyenv, etc.
      - export PATH="$HOME/.bun/bin:$PATH"
    env:
      PYTHONDONTWRITEBYTECODE: "1"

These commands are automatically prepended to every task, so your .rr.yaml tasks stay clean:

Copy# .rr.yaml - no need to repeat setup in each task
tasks:
  test:
    run: uv run pytest -v
  build:
    run: bun run build

SSH entries can be:

• Hostnames: mac-mini.local, 192.168.1.50
• User@host: deploy@server.example.com
• SSH config aliases: Names defined in ~/.ssh/config (e.g., dev-server, mac-mini-tailscale)

Passwordless SSH is required. The user must be able to run ssh <alias> without entering a password. This typically means configuring key-based auth in ~/.ssh/config.

Project Config (.rr.yaml)

Shareable project settings. Can be committed to version control. References hosts by name from global config.

Copyversion: 1

# Reference hosts from global config (optional)
hosts:
  - mini
  - server

# Defaults applied to all tasks (reduces repetition)
defaults:
  setup:
    - source ~/.local/bin/env     # Source environment (uv, pyenv, etc.)
    - set -o pipefail             # Fail on pipe errors
  env:
    PYTHONDONTWRITEBYTECODE: "1"

sync:
  exclude:
    - .git/
    - node_modules/
    - .venv/
  preserve:
    - .venv/
    - node_modules/

lock:
  enabled: true
  timeout: 5m

tasks:
  test:
    run: pytest -v      # defaults.setup runs first automatically
  build:
    run: make build

Project Defaults

The defaults section reduces repetition across tasks:

Merge order (lowest to highest precedence):

1. Host env (from global config)
2. Project defaults.env
3. Task-specific env

For setup commands:

1. Host setup_commands (from global config)
2. Project defaults.setup
3. Then the task command runs

Commands

Core Commands

Host Management

Diagnostics & Monitoring

Setup & Utilities

Common Flags

Most commands accept:

• --host <name> - Target specific host
• --tag <tag> - Select host by tag
• --probe-timeout <duration> - SSH probe timeout (e.g., 5s)

Sync-specific:

• --dry-run - Show what would sync without syncing

Variable Expansion

The dir field in host config supports:

Tasks

Define reusable commands in .rr.yaml:

Copytasks:
  test:
    description: Run all tests
    run: pytest -v

  deploy:
    description: Build and deploy
    steps:
      - name: Build
        run: make build
      - name: Deploy
        run: ./deploy.sh
        on_fail: stop

Run with: rr test, rr deploy

Extra arguments are appended to single-command tasks:

Copyrr test tests/test_api.py    # Runs: pytest -v tests/test_api.py
rr test -k "test_login"      # Runs: pytest -v -k "test_login"

Note: Args are only supported for tasks with a single run command, not multi-step tasks.

Parallel Tasks

Run multiple tasks concurrently across available hosts. Useful for running tests on multiple architectures or parallelizing independent jobs:

Copytasks:
  test-all:
    description: Run all tests in parallel
    parallel:
      - test
      - lint
      - vet
    fail_fast: false    # Continue even if one fails
    timeout: 10m

  quick-check:
    description: Fast verification
    parallel:
      - vet
      - lint
    fail_fast: true     # Stop on first failure
    max_parallel: 2     # Limit concurrency

Run with: rr test-all, rr quick-check

Parallel Task Flags

Output Modes

• progress (default): Live status indicators with spinners
• stream: Real-time output with [host:task] prefixes
• verbose: Full output shown when each task completes
• quiet: Summary only at the end

Example:

Copyrr test-all --stream    # See all output in real-time
rr test-all --dry-run   # Preview what would run
rr test-all --local     # Run locally without remote hosts

Work-Stealing Distribution

Tasks are distributed across hosts using a work-stealing queue. Fast hosts automatically grab more work, providing natural load balancing without pre-assignment.

Log Storage

Task output is saved to ~/.rr/logs/<task>-<timestamp>/:

• One file per subtask
• Summary file with timing and results

Use rr logs to view recent logs or rr logs clean to remove old ones.

Multi-Step Task Progress

Multi-step tasks show progress as each step runs:

Copy━━━ Step 1/3: Build ━━━
$ make build
[output...]
● Step 1/3: Build (2.3s)

━━━ Step 2/3: Test ━━━
$ pytest -v
[output...]
● Step 2/3: Test (45.1s)

━━━ Step 3/3: Deploy ━━━
$ ./deploy.sh
[output...]
● Step 3/3: Deploy (5.2s)

How It Works

1. Host Selection: Tries SSH aliases in order until one connects and is not busy
2. File Sync: Uses rsync with exclude/preserve patterns
3. Locking: Creates lock on remote before running; if host locked, tries next host
4. Load Balancing: When all hosts locked, round-robins until one frees up
5. Output Formatting: Auto-detects pytest/jest/go test and formats failures

Troubleshooting

Check Configuration

Copyrr doctor           # Full diagnostic
rr host list        # See configured hosts

Connection Issues

If SSH fails:

1. Verify ssh <alias> works manually
2. Check ~/.rr/config.yaml has correct SSH aliases
3. Run rr doctor for detailed diagnostics

Sync Slow

Add exclusions to .rr.yaml:

Copysync:
  exclude:
    - .git/
    - node_modules/
    - .venv/
    - __pycache__/

Stuck Lock

Copyrr unlock              # Default host
rr unlock <hostname>   # Specific host
rr unlock --all        # All hosts

Machine Interface (LLM/CI Mode)

When running rr from an LLM or CI context, use --machine (or -m) for structured JSON output with a consistent envelope format.

Global Flag

Copyrr --machine <command>   # JSON output with success/error envelope
rr -m doctor             # Short form

Read Commands with JSON Output

JSON Envelope Format

All --machine output follows this structure:

Copy{
  "success": true,
  "data": { /* command-specific output */ },
  "error": null
}

On failure:

Copy{
  "success": false,
  "data": null,
  "error": {
    "code": "SSH_AUTH_FAILED",
    "message": "Authentication failed for host m1-mini",
    "suggestion": "Run: ssh-copy-id m1-mini"
  }
}

Error Codes

Non-Interactive Commands

For CI/automation, use flag-based commands instead of interactive prompts:

Copy# Add host without prompts
rr host add --name dev-box \
  --ssh "dev.local,dev-tailscale" \
  --dir '~/projects/${PROJECT}' \
  --tag fast \
  --env "DEBUG=1" --env "PATH=/custom/bin:$PATH" \
  --skip-probe

# Initialize project without prompts
rr init --non-interactive --host dev-box

Troubleshooting Decision Tree (LLM)

Copy1. Run: rr doctor --machine
2. Parse: response.success
   - true  -> Setup OK
   - false -> Check response.error.code

3. Based on error.code:

   CONFIG_NOT_FOUND:
     -> Run: rr init --non-interactive --host <host>

   SSH_TIMEOUT:
     -> Check network: ping <hostname>
     -> Try alternate SSH alias

   SSH_AUTH_FAILED:
     -> Run: rr setup <hostname>
     -> Or: ssh-copy-id <hostname>

   SSH_HOST_KEY:
     -> Inform user about host key verification
     -> Run: ssh -o StrictHostKeyChecking=accept-new <hostname> exit

   LOCK_HELD:
     -> Run: rr unlock
     -> Retry original command

When to Use rr vs Local Execution

CopyIF .rr.yaml exists AND rr status --machine shows healthy hosts:
  -> Use rr for tests, builds, remote commands

IF no .rr.yaml OR all hosts unhealthy:
  -> Check if local_fallback is enabled in config
  -> If yes: rr will run locally automatically
  -> If no: run commands locally with Bash tool

When to Use Each Command

Example Workflows

Initial Setup

Copy# 1. Add a host to global config
rr host add

# 2. Initialize project config
cd your-project
rr init

# 3. Verify everything works
rr doctor

# 4. Run something
rr run "make test"

Daily Development

Copy# Run tests
rr run "pytest -v"

# Or use a named task
rr test

# Quick command without sync
rr exec "git status"

# Monitor hosts while working
rr monitor

Multiple Hosts (Load Balancing)

Copy# ~/.rr/config.yaml
hosts:
  gpu-1:
    ssh: [gpu1.local, gpu1-tailscale]
    dir: ~/projects/${PROJECT}
  gpu-2:
    ssh: [gpu2.local, gpu2-tailscale]
    dir: ~/projects/${PROJECT}

Copy# .rr.yaml
hosts:
  - gpu-1
  - gpu-2

Now rr run automatically uses whichever GPU box is free.