Command Cheat Sheet

Essentials

roborev init [--agent <name>]    # Initialize repo + daemon + hook
                                 # --no-daemon: skip auto-starting daemon
roborev fix                      # Fix open reviews
roborev status                   # Check daemon and queue
roborev tui                      # Interactive terminal UI
                                 # --repo: pre-filter to repo
                                 # --branch: pre-filter to branch
roborev version                  # Show version

Reviewing Code

# Single commits
roborev review                   # Review HEAD
roborev review <sha>             # Review specific commit

# Commit ranges
roborev review <start> <end>     # Review range (inclusive)
roborev review --since <commit>  # Review since commit (exclusive)
roborev review --since HEAD~5    # Review last 5 commits

# Branch reviews
roborev review --branch          # All commits since diverged from main
roborev review --branch=feature-xyz     # Review a specific branch
roborev review --branch --base develop  # Against specific base

# Uncommitted changes
roborev review --dirty           # Review working tree
roborev review --dirty --wait    # Wait for result

# Review types
roborev review --type security   # Security-focused review
roborev review --type design     # Design-focused review

Flag	Description
--wait	Wait for review to complete and display result
--quiet	Only show progress/elapsed time
--branch [name]	Review all commits on branch since base (optionally specify branch name)
--base <branch>	Base branch for --branch comparison (default: auto-detect)
--agent <name>	Use specific agent for review
-m, --model <model>	Model to use (format varies by agent)
--type <type>	Review type (security, design) — changes system prompt
--reasoning <level>	Set reasoning depth (thorough/standard/fast)
--fast	Shorthand for --reasoning fast
--local	Run review locally without the daemon (streams output to console)
--repo <path>	Specify repository path

See: Reviewing Code

Waiting for Reviews

roborev wait                     # Wait for most recent job for HEAD
roborev wait abc123              # Wait for most recent job for commit
roborev wait 42                  # Job ID (if "42" is not a valid git ref)
roborev wait 42 43 44            # Wait for multiple jobs
roborev wait --job 42            # Force as job ID
roborev wait --sha HEAD~1        # Wait for job matching HEAD~1
roborev wait --quiet             # Suppress output (for hooks/agents)

Flag	Description
--sha <ref>	Git ref to find the most recent job for
--job	Force argument to be treated as job ID
-q, --quiet	Suppress output (exit code only)

Unlike roborev review --wait, this does not enqueue a new review. It waits for an already-running job, making it useful when a post-commit hook has already triggered the review.

See: Reviewing Code

Viewing Reviews

roborev show                     # Show review for HEAD
roborev show <sha>               # Show review for commit
roborev show <job_id>            # Show review by job ID
roborev show --job <id>          # Force interpretation as job ID
roborev show --prompt <job_id>   # Show the prompt sent to the agent
roborev list                     # List jobs for current repo/branch
roborev list --open              # List only open reviews
roborev list --closed            # List only closed reviews
roborev tui                      # Interactive browser
roborev tui --repo --branch      # Pre-filtered to current repo+branch
roborev log <job_id>             # View job log

Flag	Description
--job	Force interpretation as job ID
--prompt	Show the prompt sent to the agent instead of the review output
--json	Output as JSON for machine-readable workflows

See: Terminal UI

Tip

Press l in the TUI to open the log viewer for any job (running or completed).

Job Logs

roborev log <job-id>             # Human-friendly rendered output
roborev log --raw <job-id>       # Raw NDJSON bytes
roborev log --path <job-id>      # Print the log file path

roborev log clean                # Remove logs older than 7 days
roborev log clean --days 3       # Remove logs older than 3 days

Flag	Description
--raw	Print raw NDJSON without formatting
--path	Print the log file path instead of contents

Job logs are persisted to ~/.roborev/logs/jobs/ so agent output remains available after daemon restarts. By default, roborev log renders NDJSON into compact, human-readable progress lines showing tool calls and agent text. Use --raw for the original NDJSON when scripting or debugging.

The clean subcommand removes log files older than the specified number of days (default: 7).

Commenting on Reviews

roborev comment <job_id> "message"   # Add comment with message
roborev comment <job_id>             # Opens editor
roborev close <job_id>               # Mark as closed
roborev close <job_id> --reopen      # Reopen a closed review

Flag	Description
--message, -m	Comment message (inline)
--commenter	Name of commenter
--job	Force interpretation as job ID

Note

roborev address is still accepted as an alias for roborev close.

See: Responding to Reviews

Auto-Fix Agentic Loop

roborev refine                   # Fix failed reviews on branch
roborev refine --max-iterations 5
roborev refine --since HEAD~3    # Refine specific range
roborev refine --quiet           # Show elapsed time only
roborev refine --list            # Preview what would be refined
roborev refine --all-branches    # Refine all branches with failures
roborev refine --branch feature  # Validate branch before refining

Flag	Description
--agent <name>	Use specific agent
-m, --model <model>	Model to use (format varies by agent)
--reasoning <level>	Set reasoning depth
--fast	Shorthand for --reasoning fast
--max-iterations <n>	Limit fix attempts (default: 10)
--since <commit>	Refine commits since specific commit
--branch <name>	Validate current branch before refining
--all-branches	Discover and refine all branches with failed reviews (implies --open)
--list	List failed reviews that would be refined without running
--newest-first	Process newest first (requires --all-branches or --list)
--quiet	Only show progress/elapsed time
--allow-unsafe-agents	Allow agents without sandboxing

See: Auto-Fix Agentic Loop with Refine

Fixing Reviews

roborev fix                        # Fix all open reviews on this branch
roborev fix 123                    # Fix a specific review by job ID
roborev fix 42 43 44               # Fix multiple reviews sequentially
roborev fix --batch                # Batch all open into one agent prompt
roborev fix --batch 42 43 44       # Batch specific jobs into one prompt
roborev fix --all-branches         # Fix all open across all branches
roborev fix --list                 # List open reviews without fixing

Flag	Description
--agent <name>	Use specific agent
-m, --model <model>	Model to use
--reasoning <level>	Set reasoning depth
--quiet	Suppress agent output
--open	Fix all open reviews on branch (default when no job IDs given)
--all-branches	Include open jobs from all branches (implies --open)
--batch	Concatenate multiple reviews into a single agent prompt instead of fixing one at a time
--list	List open reviews with details (job ID, ref, branch, agent, verdict) without running any fixes
--newest-first	Process jobs newest first instead of oldest first (requires --open)

See: Assisted Refactoring

Consolidating Reviews

roborev compact                              # Enqueue consolidation (background)
roborev compact --wait                       # Wait for completion
roborev compact --branch main                # Compact jobs on main branch
roborev compact --all-branches               # Compact jobs across all branches
roborev compact --dry-run                    # Show what would be done
roborev compact --limit 10                   # Process at most 10 jobs
roborev compact --agent claude-code          # Use specific agent for verification
roborev compact --reasoning thorough         # Use thorough reasoning level

Flag	Description
--wait	Block until consolidation completes
--branch <name>	Filter by branch (default: current branch)
--all-branches	Compact jobs across all branches
--dry-run	Preview what would be done without running
--limit <n>	Maximum number of jobs to process (default: 20)
--agent <name>	Agent for verification
-m, --model <model>	Model to use
--reasoning <level>	Set reasoning depth (thorough/standard/fast)
--timeout <duration>	Timeout for --wait mode (default: 10m)
--quiet	Suppress progress output

Compact discovers open completed reviews, sends them to an agent for verification against the current codebase, and consolidates related findings into a single review job. Original jobs are automatically closed when consolidation finishes. This adds a quality layer between review and fix to reduce false positives.

Note

Avoid running multiple compact commands concurrently on the same branch. The operation is not atomic and concurrent runs can produce inconsistent state.

CI Review

roborev ci review                            # Auto-detect from GitHub Actions env
roborev ci review --ref HEAD~3..HEAD         # Explicit ref range
roborev ci review --gh-repo myorg/myrepo --pr 42  # Explicit repo and PR
roborev ci review --agent codex --agent gemini     # Multiple agents
roborev ci review --comment                  # Post results as PR comment

Flag	Description
--ref <range>	Git ref or range to review (default: auto-detect from GITHUB_REF)
--comment	Post results as a PR comment via gh
--gh-repo <owner/repo>	GitHub repo (default: GITHUB_REPOSITORY env var)
--pr <number>	PR number (default: extracted from GITHUB_EVENT_PATH)
--agent <names>	Agents to use (repeatable, default: auto-detect)
--review-types <types>	Review types to run (comma-separated: security, design, default)
--reasoning <level>	Reasoning depth (thorough/standard/fast)
--min-severity <level>	Minimum severity to report (low/medium/high/critical)
--synthesis-agent <name>	Agent for combining multi-job results

Runs a one-shot review without a daemon or database. Designed for CI pipelines where you want review results as part of the build, not as a background service.

In GitHub Actions, ci review auto-detects GITHUB_REPOSITORY, GITHUB_REF, and GITHUB_EVENT_PATH so you can run it with no flags. Outside GitHub Actions, pass --gh-repo and --ref explicitly.

Exit codes: 0 on success or when all agents were skipped due to quota exhaustion, non-zero on real failures.

See: GitHub Integration

GitHub Actions Setup

roborev init gh-action                          # Generate workflow file
roborev init gh-action --agent codex            # Specify agents
roborev init gh-action --output .github/workflows/review.yml
roborev init gh-action --force                  # Overwrite existing
roborev init gh-action --roborev-version 0.34.0 # Pin version

Flag	Description
--agent <names>	Agents to include in the workflow (repeatable)
--output <path>	Output path (default: .github/workflows/roborev.yml)
--force	Overwrite an existing workflow file
--roborev-version <ver>	Pin roborev version in the workflow (default: latest)

Generates a GitHub Actions workflow that:

1. Checks out the repository
2. Downloads and installs roborev with SHA256 verification
3. Runs roborev ci review --comment on each PR
4. Posts review results as PR comments

Agent API keys are read from repository secrets (e.g. ANTHROPIC_API_KEY for Claude Code, OPENAI_API_KEY for Codex). Add the required secrets in your repository’s Settings > Secrets and variables > Actions.

See: GitHub Integration

Code Analysis

roborev analyze test-fixtures internal/*_test.go   # Run built-in analysis
roborev analyze refactor --fix ./...               # Analyze and auto-fix
roborev analyze complexity --per-file src/*.go      # One job per file
roborev analyze refactor --branch                  # Analyze changed files on branch
roborev analyze refactor --branch=feature-xyz      # Analyze a specific branch
roborev analyze --list                             # List analysis types

Flag	Description
--agent <name>	Use specific agent
-m, --model <model>	Model to use
--reasoning <level>	Set reasoning depth
--wait	Wait for completion and display result
--quiet	Suppress output
--branch [name]	Analyze files changed on branch (optionally specify branch name)
--base <branch>	Base branch for --branch comparison (default: auto-detect)
--per-file	One analysis job per file
--fix	Analyze then apply fixes automatically
--fix-agent <name>	Agent for fix step
--json	Output job info as JSON
--list	List available analysis types
--show-prompt <type>	Show prompt template

See: Assisted Refactoring

Custom Agent Tasks

roborev run "Explain the architecture"
roborev run --wait "Review src/auth/ for security issues"
roborev run "Find simplification opportunities in src/utils/"
roborev run --agentic "Add input validation to all endpoints"
cat review-checklist.txt | roborev run --wait

Flag	Description
--wait	Wait for completion and display result
--quiet	Only show progress/elapsed time
--agent <name>	Use specific agent
--reasoning <level>	Set reasoning depth
--agentic, --yolo	Enable agentic mode (can modify files)
--no-context	Don’t include repository context
--label <string>	Custom label displayed in TUI (default: run)

See: Custom Agent Tasks

Configuration

roborev config get <key>             # Get value (merged: local then global)
roborev config get <key> --global    # Get from global config only
roborev config get <key> --local     # Get from repo config only

roborev config set <key> <value>     # Set in repo config (default)
roborev config set <key> <value> --global  # Set in global config

roborev config list                  # List merged config
roborev config list --show-origin    # Show where each value comes from

Flag	Description
--global	Use global config (~/.roborev/config.toml)
--local	Use repo config (.roborev.toml)
--show-origin	Show origin column (global/local/default) in list output

See: Configuration

Repository Management

roborev repo list                # List all repos
roborev repo show <name>         # Show repo details
roborev repo rename <old> <new>  # Rename display name
roborev repo delete <name>       # Remove from tracking
roborev repo merge <src> <dst>   # Merge reviews between repos

See: Repository Management

Daemon & Hooks

roborev daemon start             # Start background daemon
roborev daemon stop              # Stop daemon
roborev daemon restart           # Restart daemon
roborev daemon run               # Run in foreground

roborev install-hook             # Install post-commit hook
roborev install-hook --force     # Overwrite existing hook with a fresh one
roborev uninstall-hook           # Remove hook

Flag	Description
--force	Overwrite an existing post-commit hook with a fresh one

Broken post-commit hook?

If your post-commit hook was corrupted during a previous upgrade (e.g. a stray fi or missing lines), run:

roborev install-hook --force

This replaces the hook entirely with a known-good version.

Checking Agents

roborev check-agents                # Smoke-test all installed agents
roborev check-agents --agent codex  # Test a specific agent
roborev check-agents --timeout 30   # Set timeout per agent (seconds)

Flag	Description
--agent <name>	Test only this agent
--timeout <secs>	Timeout per agent (default: 60)

Agent Skills

roborev skills install           # Install skills for agents
roborev skills update            # Update installed skills

See: Agent Skills

Sync & Streaming

roborev sync status              # Show PostgreSQL sync status
roborev sync now                 # Trigger immediate sync

roborev stream                   # Stream all events (JSONL)
roborev stream --repo .          # Filter to current repo

See: PostgreSQL Sync, Event Streaming

Multi-Repo Workspaces

roborev list looks in immediate child subfolders for repositories, so you can run it from a parent directory that contains multiple repos. roborev review suggests repo-level review commands when run from a workspace root, making it easy to review across projects.

Global Flags

These flags work across most commands:

Flag	Description
--server <addr>	Daemon address (default: http://127.0.0.1:7373)
-v, --verbose	Verbose output

Update

roborev update                   # Update to latest version
roborev update --force           # Force update (useful for dev builds)