v2.1.1 — Detect & Heal: The Industry's First Self-Healing Code Review

AI Code Review
That Fixes Its Own Findings

One-Click Heal

Detect issues → AI fixes → Auto-commit. Other tools just tell you what's wrong. OCR fixes it for you. Zero human intervention, end-to-end from scan to heal.

Try Self-Heal Now →View on GitHub
GitHub Stars npm downloads MIT → BSL-1.1
npm versionnpm downloadsGitHub stars

Detect → Heal Pipeline

From finding issues to fixing code, one command. Other tools just report. OCR fixes it.

Scan

Analyze AI-generated code with ocr scan

Detect

Find hallucinated packages, logic gaps, security issues

Analyze

Deep root-cause analysis with LLM reasoning

Heal

AI auto-generates fixes, dry-run preview, then apply

Clean Code

Zero human intervention, code ships clean

From finding issues to fixing code, one command

How OCR Works — 3 Steps to Clean Code

Step 1: Scan

Run ocr scan in your CI pipeline or locally. OCR analyzes your code across 3 levels (L1/L2/L3) and 6 languages.

Step 2: Detect

AI-powered detection finds:

  • •Hallucinated packages & APIs
  • •Logic gaps & dead code
  • •Security vulnerabilities
  • •AI-specific failure patterns

Step 3: Heal

Run ocr heal and let AI fix the issues:

  • •Automatic fix generation with LLM reasoning
  • •Dry-run mode to preview changes
  • •One command to patch your codebase
  • •Works with any LLM provider (GLM free!)
Result: Clean, production-ready code. Clean, production-ready code.

Add to Your CI in 30 Seconds

Works with GitHub Actions and GitLab CI

npx @opencodereview/cli@latest scan ./src
npx @opencodereview/cli@latest heal ./src

AI auto-fix detected issues — scan, heal, ship

GitHub Actions

# .github/workflows/ci.yml
- uses: raye-deng/open-code-review@v1
  with:
    threshold: 70
    paths: 'src/**/*.ts'
    fail-on-low-score: true

GitLab CI

# .gitlab-ci.yml
include:
  - component: open-code-review/validate@v1
    inputs:
      threshold: 70
      paths: src

Live Terminal Output

Real output from ocr scan and ocr heal commands

ocr scan — bash
$ ocr scan ./src --sla L3 --provider glm
╔══════════════════════════════════════════════════╗
║  Open Code Review v2.1.1 — L3 Deep Scan          ║
╚══════════════════════════════════════════════════╝
📂 Scanning 47 files across 3 languages...
🔍 L1: Structural Analysis.............. ✓ 12 issues
🔍 L2: Embedding Recall (Ollama)....... ✓ 5 issues
🤖 L3: LLM Deep Scan (GLM-4.5-Air)... ✓ 3 issues
⏱️  Total: 8.2s
┌─────────────────────────────────────────────────┐
│  Quality Score: 72/100                           │
│  ████████████████████░░░░░  72%                  │
├──────────┬──────────┬──────────┬─────────────────┤
│ Complete │ Coherent │ Consist. │ Concise         │
│   78/100 │   70/100 │   68/100 │   72/100        │
└──────────┴──────────┴──────────┴─────────────────┘
🔴 Critical (2):
  • [C-001] Hallucinated package: @ai-ui/components
    → Line 5: import { SmartTable } from '@ai-ui/components'
    → Fix: No such package exists on npm registry
  • [C-002] Undefined function: validateInput()
    → Line 23: const result = validateInput(formData)
    → Fix: Function declared but never implemented
🟡 Warning (3):
  • [W-001] Empty catch block — Line 45
  • [W-002] TODO marker without ticket — Line 67
  • [W-003] Unused import: 'lodash' — Line 2
📋 Full report: ./ocr-report.json
📊 SARIF: ./ocr-report.sarif

Quality Score Dashboard

Quantified code quality across 4 dimensions

72/100
Quality Score
L3 Deep Scan — GLM-4.5-Air
Completeness78/100
Coherence70/100
Consistency68/100
Conciseness72/100
●≥80 Good●60-79 Fair●<60 Poor

See OCR in Action

Before: AI-generated code with defects → After: OCR auto-fixed, zero defects

❌ Before — scan
// ❌ AI-generated code — 3 defects found
import { SmartForm } from '@ai-ui/smart-form';  // 🚨 hallucinated
import _ from 'lodash';                            // ⚠️ unused
function handleSubmit(data: any) {
  try {
    const result = processData(data);
    return result;
  } catch (e) {                                   // ⚠️ empty catch
    // TODO: handle error later
  }
}
ocr heal
— 2.1s —
✅ After — heal
// ✅ OCR auto-fixed — 0 defects
import { Form } from 'antd';
async function handleSubmit(data: FormData): Promise<void> {
  try {
    const result = await processData(data);
    return result;
  } catch (error) {
    logger.error('Submit failed:', error);
    throw new SubmitError('Form submission failed');
  }
}

Why Existing Tools Aren't Enough

ESLint, SonarQube, and CodeClimate were built for human-written code. AI-generated code has fundamentally different failure modes.

Traditional Linters

  • ✗Can't detect hallucinated npm packages
  • ✗Miss logic gaps from context window limits
  • ✗Don't understand AI code generation patterns
  • ✗No feedback loop to AI assistants
  • ✗Style-focused, not logic-focused

Open Code Review — Detect & Heal

  • ✓Detects phantom packages, functions, and APIs
  • ✓Catches logic discontinuities and dead code
  • ✓Purpose-built for AI code failure modes
  • ✓Self-heal: auto-fixes, not just reports
  • ✓0-100 quality score with dimensional breakdown

Core Capabilities

Detect & Heal — The Industry's First Self-Healing Code Review

🩹

Detect & Heal — Self-Healing Code Review

`ocr scan` detects AI code defects, `ocr heal` auto-generates fixes. Supports dry-run preview → confirm → auto-fix. Works with 8 LLM Providers (GLM/DeepSeek/OpenAI/Ollama etc.). One command, from detection to fix, zero human intervention.

👁️

Hallucination Detection

Catches phantom packages, undefined functions, and non-existent APIs that AI models confidently generate.

🧩

Logic Gap Analysis

Identifies empty catch blocks, unreachable code, TODO markers, and missing error handling from context window limits.

👻

Phantom Package DB

Real-time database of hallucinated npm/PyPI packages. Detects AI-fabricated dependencies before they break production.

📊

Quality Score (0-100)

Quantified scoring across 4 dimensions: completeness, coherence, consistency, and conciseness.

🚀

CI/CD Quality Gate

Block low-quality AI code from merging. Works with GitHub Actions and GitLab CI Components. SARIF output supported.

🔬

L3 Deep Scan — Foundation for Precise Fixes

Deep detection is the foundation for precise fixes. Suspicious code blocks sent to remote LLMs for thorough analysis. Supports 8 providers: OpenAI, GLM, DeepSeek, Ollama, and more. GLM/Ollama are free!

🌍

Multi-Language AI Detection

6 language-specific detectors: TypeScript/JavaScript, Python, Java, Go, Kotlin, Rust. Each language has its own AI defect pattern recognition for precise detection.

🔗

Universal Provider Adapter

OpenAI-compatible adapter for any LLM service. Built-in provider presets auto-fill baseUrl. Switch with --provider, --api-key, --model flags.

🎯

3-Tier SLA Scanning

L1 Fast, L2 Standard (Embedding + Local LLM), L3 Deep (Embedding + Remote LLM) — choose the depth your project needs. L3 is free for individuals (bring your own API key).

Real Teams, Real Results

See how teams use Open Code Review to ship AI-generated code with confidence.

AI Startup

AI Startup — Automated PR Review

Team uses Cursor + Copilot for daily coding. Every PR automatically runs ocr scan + ocr heal before merge.

✓Low-score code is auto-blocked before merge. AI auto-fixes common hallucinated packages and logic gaps. 70% less review time.

After using OCR, our PR review time dropped by 70%.

— CTO, AI SaaS Startup

GitHub ActionsCursorTypeScript
🏦Enterprise

FinTech — AI Code Compliance

Regulations require all AI-generated code to go through review. OCR runs in CI to automatically scan every merge request.

✓Every MR generates a quality report. SARIF output integrates with GitLab Code Quality. Passed compliance audit on first attempt.

OCR helped us pass our compliance audit on the first try.

— VP Engineering, FinTech Company

GitLab CIJavaPython
🐙Open Source

Open Source — Contribution Review

Maintainers use OCR to review community contributions. Automatically detects hallucinated packages and AI-generated code with defects.

✓Zero manual effort for contribution quality review. Suspicious PRs are auto-flagged. Maintainers only review genuine issues.

No more worrying about contributors submitting hallucinated code written by ChatGPT.

— Maintainer, 5k★ Open Source Project

GitHub ActionsGoOpenAI

Free for Individuals

Full CLI + AI Self-Heal, $0 forever
Team plans from $19/seat/month

View Pricing →

Frequently Asked Questions

What is Open Code Review?

The industry's first self-healing AI code review tool. Detect hallucinations, logic gaps, then auto-fix them with AI. End-to-end from scan to heal, zero human intervention.

Is it really free?

Yes. The CLI and all local analysis features are free forever. GLM/Ollama can also be used for free with the self-heal feature. No credit card, no trial limits.

How is it different from ESLint?

Traditional linters just check code style. Open Code Review detects AI-specific issues — hallucinated packages, logic gaps — and doesn't just report them, it fixes them.

Which CI/CD platforms are supported?

GitHub Actions and GitLab CI Components are officially supported. Run locally via CLI too.

Does it work with Cursor or Copilot?

Yes. ocr heal auto-fixes code and generates IDE rule files so AI assistants write better code from the start.

What is Detect & Heal?

OCR's core differentiator: ocr scan detects issues, ocr heal auto-fixes them with AI. Other tools just tell you what's wrong — OCR fixes it for you. Supports 8 LLM providers, GLM is completely free.