Alibaba Open-Sources Its Internal AI Code Review CLI

Alibaba has released Open Code Review (OCR), a command-line tool that wraps an LLM-driven agent around Git diffs to produce line-level review comments. The project grew out of an internal assistant used across the company for two years, and now ships as an npm package or prebuilt binary that points at any configurable model endpoint, including Anthropic-compatible APIs and Claude Code’s environment variables.

The pitch is a deliberate split between deterministic plumbing and agent reasoning. Hard-coded engineering handles file selection, bundling related files (like paired i18n resource files) into isolated sub-agent contexts, matching rules to file types via a template engine, and running separate positioning and reflection passes to anchor comments to the right lines. The agent itself is reserved for dynamic decisions and context retrieval, using prompts and a toolset tuned from production call traces. Alibaba frames this as a fix for the failure modes of generic agents like Claude Code Skills: skipped files on large changesets, drifting line numbers, and prompt-sensitive quality.

Integration paths include a Claude Code plugin, a slash command, and GitHub Actions / GitLab CI examples that consume JSON output for PR automation. Concurrency flags and preview mode round out the CLI for larger repositories.