OpenCodeReview Code Review Skill
Precise Git and full-file review through deterministic pipelines and a configurable LLM agent.
What does this skill do, and when should you use it?
OpenCodeReview is Alibaba’s open-source AI code-review CLI, originating from an internal assistant used at large scale. It reads Git diffs, sends changed files to a configurable LLM agent, and produces structured comments with line-level locations. Deterministic file selection, bundling, rule matching, positioning, and reflection modules constrain the review workflow, while the agent retrieves broader repository context. It supports workspace, branch, commit, and full-file reviews, plus Claude Code, Codex, Cursor, and CI/CD integrations.
Reads workspace changes, commits, or branch ranges; filters and bundles files for review; and calls a configured Anthropic- or OpenAI-compatible LLM endpoint. The agent can read complete files, search the repository, and inspect related changed files. The CLI produces comments with file locations, line numbers, categories, and severities, with text or JSON output. ocr scan reviews complete repositories, directories, or files and can provide project summaries, deduplication, and token budgets.
- Development teams reviewing staged, unstaged, and untracked changes before committing or merging.
- Engineering teams parsing JSON findings in CI/CD to sort, filter, group, or gate builds.
- Maintainers auditing unfamiliar repositories, migration targets, or directories without meaningful diffs.
- Developers using Claude Code, Codex, or Cursor who want an agent skill or plugin to invoke a local review CLI.
- Teams applying built-in or custom checks for issues such as null safety, thread safety, XSS, and SQL injection.
What are this skill's strengths and limitations?
- Covers workspace changes, branch ranges, commits, and full-file scans.
- Uses deterministic selection, rule matching, and comment-positioning stages to improve consistency.
- Supports Anthropic, OpenAI-compatible endpoints, and custom providers.
- Offers structured JSON output, a GitHub Action, CI examples, and resumable sessions.
- Supports MCP servers and custom review rules.
- Requires the CLI, Git, and a configured LLM endpoint, with potential model-service costs.
- The README explicitly describes lower Recall than general-purpose agents as a deliberate precision-over-noise trade-off.
- The SKILL.md body is not included in the supplied source, so its complete instruction set and compatibility details cannot be verified.
- The supplied source does not provide a test-suite inventory, performance limits, or validation evidence for every target platform.
How do you install this skill?
Install Git >= 2.41. Install the CLI with npm install -g @alibaba-group/open-code-review, or on macOS/Linux run curl -fsSL https://raw.githubusercontent.com/alibaba/open-code-review/main/install.sh | sh. Install the skill with npx skills add alibaba/open-code-review --skill open-code-review. From source, clone the repository, run make build, and place dist/opencodereview on PATH. The README does not document the exact folder layout for the SKILL.md installation.
How do you use this skill?
Run ocr config provider and ocr config model to configure an LLM, then test it with ocr llm test. Review current changes with ocr review, compare branches with ocr review --from main --to feature-branch, or scan a repository with ocr scan. In a supported coding agent, invoke @Open Code Review review my current changes, or use ocr review --audience agent.
How does this skill compare with similar options?
Compared with the general-purpose agents named in the README, such as Claude Code with Skills, OpenCodeReview adds deterministic controls for file selection, rule matching, and comment positioning. Its stated trade-off is lower Recall in exchange for higher Precision, less noise, and lower token use.