reviewcerberus 1.0.2

AI-powered code review tool for git branches

ReviewCerberus

AI-powered code review tool that analyzes git branch differences and generates comprehensive review reports with structured output.

Key Features

• Comprehensive Reviews: Detailed analysis of logic, security, performance, and code quality
• Structured Output: Issues organized by severity with summary table
• Multi-Provider: AWS Bedrock, Anthropic API, or Ollama
• Smart Analysis: Context provided upfront with prompt caching
• Git Integration: Works with any repository, supports commit hashes

---

Quick Start

Run with Docker (recommended):

docker run --rm -it -v $(pwd):/repo \
  -e MODEL_PROVIDER=anthropic \
  -e ANTHROPIC_API_KEY=sk-ant-your-api-key \
  kirill89/reviewcerberus-cli:latest \
  --repo-path /repo --output /repo/review.md

That's it! The review will be saved to review.md in your current directory.

See Configuration for AWS Bedrock setup and other options.

---

Usage

Basic Commands

# Run code review
poetry run reviewcerberus

# Custom target branch
poetry run reviewcerberus --target-branch develop

# Custom output location
poetry run reviewcerberus --output /path/to/review.md
poetry run reviewcerberus --output /path/to/dir/  # Auto-generates filename

# Different repository
poetry run reviewcerberus --repo-path /path/to/repo

# Add custom review guidelines
poetry run reviewcerberus --instructions guidelines.md

Example Commands

# Full review with custom guidelines
poetry run reviewcerberus --target-branch main \
  --output review.md --instructions guidelines.md

# Review a different repo
poetry run reviewcerberus --repo-path /other/repo

---

What's Included

Comprehensive Code Review

Detailed analysis covering:

• Logic & Correctness: Bugs, edge cases, error handling
• Security: OWASP issues, access control, input validation
• Performance: N+1 queries, bottlenecks, scalability
• Code Quality: Duplication, complexity, maintainability
• Side Effects: Impact on other system parts
• Testing: Coverage gaps, missing test cases
• Documentation: Missing or outdated docs, unclear comments

Structured Output

Every review includes:

• Summary: High-level overview of changes and risky areas
• Issues Table: All issues at a glance with severity indicators (🔴 CRITICAL, 🟠 HIGH, 🟡 MEDIUM, 🟢 LOW)
• Detailed Issues: Each issue with explanation, location, and suggested fix

---

How It Works

1. Detects current git branch and repository
2. Collects all context upfront: changed files, commit messages, and diffs
3. Analyzes using AI agent with access to:
   • Full diff context (truncated at 10k chars per file)
   • File reading with line ranges
   • Pattern search across codebase
   • Directory listing
4. Generates structured review output rendered as markdown

Progress Display:

Repository: /path/to/repo
Current branch: feature-branch
Target branch: main

Found 3 changed files:
  - src/main.py (modified)
  - src/utils.py (modified)
  - tests/test_main.py (added)

Starting code review...

🤔 Thinking... ⏱️  3.0s
🔧 read_file_part: src/main.py

✓ Review completed: review_feature-branch.md

Token Usage:
  Input tokens:  6,856
  Output tokens: 1,989
  Total tokens:  8,597

---

Configuration

All configuration via environment variables (.env file):

Provider Selection

MODEL_PROVIDER=bedrock  # or "anthropic" or "ollama" (default: bedrock)

AWS Bedrock (if MODEL_PROVIDER=bedrock)

AWS_ACCESS_KEY_ID=your_key
AWS_SECRET_ACCESS_KEY=your_secret
AWS_REGION_NAME=us-east-1
MODEL_NAME=us.anthropic.claude-opus-4-5-20251101-v1:0  # optional

Docker example with Bedrock:

docker run --rm -it -v $(pwd):/repo \
  -e AWS_ACCESS_KEY_ID=your_key \
  -e AWS_SECRET_ACCESS_KEY=your_secret \
  -e AWS_REGION_NAME=us-east-1 \
  kirill89/reviewcerberus-cli:latest \
  --repo-path /repo --output /repo/review.md

Anthropic API (if MODEL_PROVIDER=anthropic)

ANTHROPIC_API_KEY=sk-ant-your-api-key-here
MODEL_NAME=claude-opus-4-5-20251101  # optional

Ollama (if MODEL_PROVIDER=ollama)

MODEL_PROVIDER=ollama
OLLAMA_BASE_URL=http://localhost:11434  # optional, default
MODEL_NAME=deepseek-v3.1:671b-cloud     # optional

Docker example with Ollama:

# Assumes Ollama running on host machine
docker run --rm -it -v $(pwd):/repo \
  -e MODEL_PROVIDER=ollama \
  -e OLLAMA_BASE_URL=http://host.docker.internal:11434 \
  kirill89/reviewcerberus-cli:latest \
  --repo-path /repo --output /repo/review.md

Optional Settings

MAX_OUTPUT_TOKENS=10000     # Maximum tokens in response
RECURSION_LIMIT=200         # Agent recursion limit

Custom Review Prompts

Customize prompts in src/agent/prompts/:

• full_review.md - Main review prompt
• context_summary.md - Context compaction for large PRs

---

Development

Local Installation

For local development (not required for Docker usage):

# Clone and install
git clone <repo-url>
poetry install

# Configure credentials
cp .env.example .env
# Edit .env with your provider credentials

See Configuration for credential setup.

Run Tests

make test
# or
poetry run pytest -v

Linting & Formatting

make lint     # Check with mypy, isort, black, mdformat
make format   # Auto-format with isort and black

Building Docker Image

make docker-build           # Build locally
make docker-build-push      # Build and push (multi-platform)

Version is auto-read from pyproject.toml. See DOCKER.md for details.

Project Structure

src/
├── config.py                        # Configuration
├── main.py                          # CLI entry point
└── agent/
    ├── agent.py                     # Agent setup
    ├── model.py                     # Model initialization
    ├── runner.py                    # Review execution
    ├── prompts/                     # Review prompts
    ├── schema.py                    # Data models (including structured output)
    ├── git_utils/                   # Git operations (changed files, diffs, commits)
    ├── formatting/                  # Context building and output rendering
    ├── progress_callback_handler.py # Progress display
    └── tools/                       # 3 review tools (read, search, list)

Code Quality Standards

• Strict type checking: All functions require type annotations
• Return types: Must be explicit (warn_return_any = true)
• Formatting: Black + isort with black profile
• Testing: Integration tests with real git operations

---

Requirements

• Python 3.11+
• Git
• One of:
  • AWS Bedrock access with Claude models
  • Anthropic API key
• Poetry (for development)

---

License

MIT