open-code-review

Performs AI-powered code review on Git changes using the ocr CLI from alibaba/open-code-review. It reads Git diffs and generates structured, line-level review comments, and can automatically apply fixes when requested. Use it when asked to review code, review a pull request, review staged or unstaged changes, review a commit, or compare branches for quality issues such as bugs, security vulnerabilities, performance problems, and code quality concerns.

Before reviewing, it verifies the environment: confirm the ocr CLI is installed (installing via npm install of @alibaba-group/open-code-review if not) and confirm LLM connectivity with the ocr llm test command. A working LLM, either Anthropic or OpenAI-compatible, must be configured first, either through environment variables (OCR_LLM_URL, OCR_LLM_TOKEN, OCR_LLM_MODEL, OCR_USE_ANTHROPIC) or persistent ocr config entries; credentials must come from the user and are never invented or hardcoded.

The workflow gathers concise business context and passes it via --background, then runs ocr review with --audience agent, which suppresses progress UI and emits only the final summary. Targets can be the default workspace mode (staged, unstaged, and untracked changes), a single commit via --commit, or a branch comparison via --from and --to, with --preview to dry-run and --timeout and --concurrency to tune runs. Each returned comment (path, content, line range, optional suggestion and reasoning) is classified High, Medium, or Low and reported grouped by priority; when line positioning fails with both lines at 0, the skill locates the code manually. Fixes focus on High and Medium items and require permission unless explicitly requested. Project-specific rules resolve from --rule, a repo or home rule.json, or built-in defaults. It is licensed Apache-2.0.