open-code-review

Open Code Review

A skill for invoking open-code-review (ocr) — an open-source AI code review CLI that reads Git diffs and generates structured, line-level review comments.

Workflow

Step 1: Gather Business Context

Analyze the review target (commits, branch, or changes) to extract concise business context. Pass this context via --background to improve review quality.

Step 2: Run Code Review

Run the OCR command with appropriate flags. Always pass business context via --background when available:

ocr review --audience agent --background "business context here" [user-args]

Argument handling:

• Background context (RECOMMENDED): use --background "context" or -b "context" to provide business context for better review quality
• Default (no user arguments): reviews staged, unstaged, and untracked changes (workspace mode)
• Specific commit: use --commit or -c to review a single commit against its parent
• Branch comparison: use --from <ref> and --to <ref> to review diff between two refs
• Timeout: default timeout is 10 minutes per file; adjust with --timeout <minutes>
• Concurrency: default concurrency is 8 file workers; reduce with --concurrency <n> if rate limits are hit
• Preview mode: use --preview or -p to preview which files will be reviewed without running the LLM
• Installation: if ocr command is not found, install it by running npm i -g @alibaba-group/open-code-review

Common invocation patterns:

Output mode:

• Always use --audience agent to suppress progress UI and emit only the final summary
• Prevent output truncation: For large reviews or restricted tool environments, redirect output to a temporary file (ocr review --audience agent ... > /tmp/ocr_out.txt 2>&1) and inspect it in full via a file reading tool instead of piping through tail or head, which drops earlier review comments.

On failure: If ocr review exits non-zero (e.g. an LLM connection error), do not retry blindly — consult the Troubleshooting section below for the matching fix before re-running.

Step 3: Report

OCR output includes structured severity (critical / high / medium / low) and category (bug / security / performance / maintainability / test / style / documentation / other) on each comment. Present results grouped by severity, discarding low severity items that are likely false positives or nitpicks.

Step 4: Fix

Before applying fixes, check whether the user requested automatic fixes:

• If the user explicitly requested "review and fix" or similar, proceed with automatic fixes
• If the user only requested "review" without fix intent, ask for permission before applying any changes

When fixing issues and suggestions:

• Focus on critical, high, and medium severity items
• Apply fixes directly to the code when safe and well-defined
• For complex fixes requiring manual intervention, clearly describe what needs to be done
• Always verify fixes with the user before committing

Output Format

Each comment in OCR's output contains:

• path: File path
• content: Review comment text
• start_line / end_line: Line range (both 0 means positioning failed)
• category: Issue category (bug, security, performance, maintainability, test, style, documentation, other)
• severity: Issue severity (critical, high, medium, low)
• suggestion_code: Optional fix suggestion
• existing_code: Optional original code snippet
• thinking: Optional LLM reasoning process

Present results grouped by severity using this template:

## Code Review Results

**Files reviewed**: N
**Issues found**: X critical, Y high, Z medium

### Critical

- **`path/to/file.java:42`** [bug] — Brief description
  > Recommendation: How to fix

### High

- **`path/to/file.java:26`** [bug] — Brief description
  > Recommendation: How to fix

### Medium

- **`path/to/file.ts:88`** [performance] — Brief description
  > Recommendation: How to fix (if applicable)

If no critical, high, or medium severity issues remain after filtering, state: "Review complete — no critical, high, or medium issues found in N files."

Handling mispositioned comments:

When start_line and end_line are both 0, the comment failed to locate the exact position in the file. In such cases:

1. Read the comment content to understand the issue
2. Examine the target file mentioned in the comment
3. Identify the relevant code section based on the comment's context
4. Apply the fix or suggestion to the correct location

Custom Review Rules

If the user wants project-specific rules, OCR resolves them in this priority order:

1. --rule <path> flag (highest)
2. <repo>/.opencodereview/rule.json
3. ~/.opencodereview/rule.json
4. Built-in system defaults (lowest)

By default, the first matching user rule replaces the built-in system rule. Set merge_system_rule: true on a rule entry when the matched system rule and user rule should both be included.

Rule file format:

{
  "rules": [
    {
      "path": "**/*.java",
      "rule": "All new methods must validate required parameters for null",
      "merge_system_rule": true
    },
    {
      "path": "**/*mapper*.xml",
      "rule": "Check SQL for injection risks and missing closing tags"
    }
  ]
}

To preview which rule applies to a file before reviewing:

ocr rules check src/main/java/com/example/Foo.java

Gotchas

• LLM must be configured first — ocr review will fail loudly if no LLM is reachable. See the Troubleshooting section below if this happens.
• Working directory matters — ocr review operates on the Git repo at the current directory. Use --repo /path/to/repo to run from elsewhere.
• Untracked files are reviewed in workspace mode — running bare ocr review includes staged, unstaged, and untracked changes. Stage selectively if you want narrower scope.
• Large diffs may hit token limits — files with very large diffs may be truncated. The default MAX_TOKENS is 58888 per request.
• Plan phase triggers at 50 lines — diffs exceeding 50 changed lines run an extra risk-analysis phase before main review. This adds latency but improves quality.
• Don't pass --audience human — it streams progress UI that pollutes output. Always use --audience agent.
• Comment language follows config — set language config to English or Chinese (default: Chinese) to control review comment language.
• Avoid output truncation — Large review runs produce verbose output. Never pipe command output to tail or head as it drops review comments from earlier sections. Redirect output to a file and read it in full.

Validation

After the review completes, verify success by checking:

1. The command exited with code 0
2. Comments were generated (or "No comments generated" message appears)
3. Warnings (if any) are displayed in stderr

If errors occurred, check the stderr warnings for details about which files failed and why.

Troubleshooting

ocr: command not found

Install the CLI:

npm install -g @alibaba-group/open-code-review

ocr review fails with LLM connection error

Prompt the user to configure an LLM provider.

Interactive setup (recommended):

ocr config provider

Manual setup (alternative):

ocr config set llm.url https://api.anthropic.com/v1/messages
ocr config set llm.auth_token <api-key>
ocr config set llm.model claude-opus-4-6
ocr config set llm.use_anthropic true

Verify connectivity with ocr llm test. Stop here and ask the user to provide credentials — never invent or hardcode API keys.

References

• Full docs: https://github.com/alibaba/open-code-review
• NPM package: https://www.npmjs.com/package/@alibaba-group/open-code-review
• Issue tracker: https://github.com/alibaba/open-code-review/issues