@open-code-review/cli

Command-line interface for Open Code Review — OpenCode-only setup, process-agent review execution, real-time progress tracking, and environment health checks.

Quick Start

Requires Node.js >= 22.5 — OCR's storage engine is Node's built-in SQLite (node:sqlite), so there's no native module to compile and it installs cleanly under any package manager (npm, pnpm 10+, yarn).

# 1. Install globally
npm install -g @open-code-review/cli

# 2. Initialize in your project
cd your-project
ocr init

ocr init configures OCR for OpenCode. Then run a review from OpenCode:

/ocr:review
/ocr-review against spec.md                          # With requirements context
/ocr-review --team architect:1,java-spring-engineer:1 # Custom team composition
/ocr-review --reviewer "Focus on error handling"     # Add an ephemeral reviewer
/ocr-map                                             # Code Review Map for large changesets

Run ocr doctor to verify your setup at any time.

Reviewer Library

OCR ships with 7 reviewer personas focused on backend service review: Architect, Database Administrator, Integration API Contract Engineer, Java Spring Engineer, Performance, QA Test Engineer, and Security Engineer.

Add ephemeral one-off reviewers with --reviewer, or create persistent custom reviewers with /ocr:create-reviewer.

Multi-Model Teams

Different reviewers can run on different models. Pair a fast generalist with a deeper specialist, or mix vendors across a single team:

# .ocr/config.yaml
models:
  aliases:
    workhorse: claude-sonnet-4-6
    big-brain: claude-opus-4-7
  default: claude-sonnet-4-6

default_team:
  architect:                            # per-instance models
    - { model: big-brain }
    - { model: workhorse, name: architect-balanced }
  security-engineer: { count: 1, model: big-brain } # one model, multiple instances
  java-spring-engineer: 2               # default model

pipeline_agents:
  aggregation: { model: big-brain }
  validation: { model: big-brain }
  synthesis: { model: big-brain }
  translation_uk: { model: big-brain }

Override per-review via --team on the CLI. OCR can list the models your installed OpenCode runtime exposes.

OpenCode reviews use ocr review run-agents internally. The CLI starts every resolved reviewer process, writes exact prompt snapshots under rounds/round-{n}/prompts/, writes reviewer stdout under reviews/, and then runs aggregation, validation, and synthesis as separate process agents. After final.md is written, it runs translation_uk and writes final.uk.md as a Ukrainian human-readable copy; translation failure is logged without invalidating the completed English review. If any configured reviewer or pipeline model is missing or any process fails, the round stops instead of continuing with a partial team or fabricated output. For normal OpenCode use, run ocr review run-agents --fresh without --background; the command keeps the shell call active, prints progress, and returns only when the review is completed or failed. If a parent tool regains control while the review is still running, use ocr review wait --session-id <id> to block until completion instead of returning a mid-run summary.

Commands

ocr init

Initialize Open Code Review in your project. Creates .ocr/ with skills, commands, and config, then configures OpenCode integration.

ocr init
ocr init --tools opencode

ocr progress

Watch a review or map session in real-time. Shows current phase, elapsed time, reviewer status, finding counts, and completion percentage.

ocr progress                           # Auto-detect current session
ocr progress --session 2026-01-26-main # Specific session

ocr doctor

Verify your OCR installation and all dependencies.

ocr doctor

Checks: git, opencode, gh (GitHub CLI), .ocr/ setup, and capabilities.

ocr update

Update OCR skills and commands after upgrading the package. Preserves your .ocr/config.yaml and all reviewer personas.

ocr update                    # Update everything
ocr update --dry-run          # Preview changes
ocr update --commands         # Commands only
ocr update --skills           # Skills and references only
ocr update --inject           # AGENTS.md/CLAUDE.md only

ocr reviewers

Manage reviewer personas. Sync metadata from reviewer markdown files to reviewers-meta.json for CLI/OpenCode reviewer resolution and generated OCR assets.

ocr reviewers sync --stdin      # Pipe JSON from AI-generated metadata

ocr state

Internal command used by the review workflow to manage session state. Subcommands: begin, advance, complete-round, complete-map, finish, status, show, sync, reconcile.

ocr state begin --session-id <id> --branch <branch> --workflow-type review
ocr state advance --phase reviews   # Advance to a phase (graph-validated)
ocr state complete-round --stdin    # Finalize a round from piped JSON
ocr state finish                    # Close the workflow (invariant-checked)
ocr state status --json             # Completeness + next action
ocr state show --json               # Show current session state
ocr state sync                      # Rebuild state from filesystem

ocr review run-agents

Execution entrypoint used for OpenCode reviews. It is normally called by the installed /ocr-review command, but can be run directly for debugging.

ocr review run-agents --session-id <session-id> --round <n>
ocr review run-agents [target] --fresh --requirements <text-or-path> --team <json>
ocr review status --session-id <session-id>
ocr review watch --session-id <session-id>
ocr review wait --session-id <session-id>

This path owns process spawning and artifact writing. Agents return stdout and NDJSON only; they do not write OCR session files directly.

Supported Runtime

This fork supports one runtime:

| Tool | Config Directory | |------|------------------| | OpenCode | .opencode/ |

Updating

After upgrading the package:

npm i -g @open-code-review/cli@latest
ocr update

The CLI notifies you when a new version is available.

Links

• Full documentation: github.com/spencermarx/open-code-review
• npm: @open-code-review/cli

License

Apache-2.0