alibaba/open-code-review

What it solves

Open Code Review addresses the instability and inaccuracy of general-purpose AI agents when performing code reviews. It solves common issues such as incomplete coverage of large changesets, "position drift" (where AI comments don't align with the actual code lines), and fluctuating review quality caused by purely natural-language prompts.

How it works

The tool uses a hybrid architecture that combines deterministic engineering with an AI agent:

• Deterministic Engineering: Handles the rigid parts of the process. It precisely selects and bundles related files, matches specific review rules to file characteristics using a template engine, and employs independent modules for comment positioning and reflection to ensure accuracy.
• AI Agent: Handles dynamic decision-making and context retrieval. It uses scenario-tuned prompts and a specialized toolset distilled from production data to perform the actual analysis of the code.

Users can run it as a CLI tool to review Git diffs or scan entire files, or integrate it into existing AI coding agents (like Claude Code or Cursor) via plugins.

Who it’s for

It is designed for developers and engineering teams who want high-precision, automated AI code reviews integrated into their local workflow or CI/CD pipelines.

Highlights

• High Precision: Prioritizes reducing false alarms over maximum recall to minimize developer noise.
• Agentic Capabilities: The agent can read full files, search the codebase, and inspect other changed files for context.
• Flexible Execution: Supports both OCR-managed LLM reviews and "Delegation Mode," where a user's own coding agent performs the review.
• Broad Integration: Compatible with multiple AI agents (Claude Code, Cursor, Codex) and CI/CD platforms (GitHub Actions, GitLab CI).
• Efficient: Claims significantly lower token consumption and faster completion times compared to general-purpose agents.