Alibaba's open-code-review claims higher precision at ~1/9 the tokens by keeping the LLM out of the correctness path.

alibaba/open-code-review is 67 days old at 12,587 stars, +1,066 today, #2 on daily trending. The Go tool splits review into a deterministic engineering pipeline (file selection, comment positioning) and an LLM agent (dynamic decisions, context retrieval), on the explicit argument that "deterministic engineering logic, not the language model, guarantees correctness." Benchmarks claim significantly higher precision and F1 than general-purpose agents at roughly one-ninth the tokens. Apache-2.0, with CI integrations for GitHub Actions, GitLab CI, GitFlic CI and Gerrit, plus Claude Code, Codex and Cursor adapters. The architectural split is the lesson: put the model where judgment is needed and nowhere else.