@open-code-review/agents

The skill definitions, reviewer personas, and workflow references that power Open Code Review.

What This Package Contains

agents/
├── skills/ocr/              # The OCR skill
│   ├── SKILL.md             # Tech Lead orchestration logic
│   ├── references/
│   │   ├── workflow.md        # 8-phase review workflow
│   │   ├── session-files.md   # Authoritative file manifest
│   │   ├── session-state.md   # State management
│   │   ├── discourse.md       # Multi-agent debate rules
│   │   ├── final-template.md  # Final review template
│   │   └── reviewers/         # Persona definitions (customizable)
│   │       ├── principal.md # Architecture, design patterns
│   │       ├── quality.md   # Code style, best practices
│   │       ├── security.md  # Auth, data handling, vulnerabilities
│   │       └── testing.md   # Coverage, edge cases
│   └── assets/
│       ├── config.yaml      # Default configuration
│       └── reviewer-template.md
├── commands/                # Slash command definitions
│   ├── review.md
│   ├── doctor.md
│   ├── history.md
│   ├── show.md
│   ├── reviewers.md
│   └── post.md
└── .claude-plugin/          # Claude Code plugin manifest
    └── plugin.json

Installation

Via CLI (Recommended)

The CLI copies these assets to your project's .ocr/ directory:

npx @open-code-review/cli init

To update after a package upgrade:

ocr update

This updates skills and workflow references while preserving your .ocr/config.yaml and all reviewer personas (both default and custom).

Via Claude Code Plugin

/plugin marketplace add spencermarx/open-code-review
/plugin install ocr@aclarify

Skill Architecture

The SKILL.md file defines the Tech Lead role—the orchestrator that:

1. Discovers project context (config, OpenSpec, reference files)
2. Analyzes changes and identifies risk areas
3. Selects and spawns reviewer personas based on your team configuration
4. Facilitates discourse between reviewers
5. Synthesizes findings into a unified review

Each reviewer in references/reviewers/ is a specialized persona. You can customize the built-in reviewers or add your own.

Session Structure

OCR uses a round-first architecture for session storage:

.ocr/sessions/{YYYY-MM-DD}-{branch}/
├── state.json              # Workflow state (current_round, phase)
├── discovered-standards.md # Project context (shared across rounds)
├── context.md              # Change analysis (shared)
└── rounds/
    ├── round-1/
    │   ├── reviews/        # Individual reviewer outputs
    │   ├── discourse.md    # Cross-reviewer discussion
    │   └── final.md        # Synthesized review
    └── round-2/            # Created on re-review
        └── ...

Multi-round reviews: Running /ocr-review again on an existing session creates a new round (round-2/, round-3/, etc.) if the previous round is complete. This enables iterative "review → fix → re-review" workflows while preserving history.

See references/session-files.md for the complete file manifest.

Commands

Why two formats? Windsurf requires flat command files with a prefix (/ocr-command), while Claude Code and Cursor support subdirectories (/ocr:command). Both invoke the same underlying functionality.

License

Apache-2.0