PR Review CLI

An AI-powered code review tool that analyzes your git diff and routes it through the right level of LLM scrutiny.

How it works

git diff → LLM classifies change size
              │
              ├── Small change ──→ Focused review (old vs new code)
              │
              └── Big change  ──→ AST analysis finds all callers
                                   └──→ Deep review (diff + impacted files)

Install

# Install the OpenAI client
pip install openai python-dotenv

# Run directly
python reviewer.py

# Or install as a CLI command
pip install -e .
git-ai-pr-review

Usage

# Set your API key
export OPENAI_API_KEY=sk-...    # macOS / Linux
setx OPENAI_API_KEY "sk-..."   # Windows (PowerShell)

# Review changes vs HEAD (default — unstaged/staged changes)
git-ai-pr-review

# Review feature branch against main
git-ai-pr-review --base main

# Review against remote main
git-ai-pr-review --base origin/main

# Review against a specific commit
git-ai-pr-review --base abc1234

Output

• Prints the review to stdout
• Saves a pr_review_output.md file in the current directory

What the LLM looks for

Small change review:

• Summary of what changed
• Bugs & edge cases
• Code quality & naming
• Concrete improvement suggestions
• Final verdict (approve / approve with suggestions / request changes)

Big change review (+ AST impact analysis):

• Everything above, plus:
• Breaking change detection (are callers still compatible?)
• Per-file impact analysis of all call sites found via AST
• Migration notes — what else needs updating

Notes

• AST analysis is Python-only (uses the ast stdlib module)
• Skips .venv and node_modules directories
• The diff-size LLM call uses OpenAI (model example: gpt-4o-mini)
• The codebase was refactored: core modules now include llm/openai.py, ast_analyzer.py, diff_parser.py, git.py, and cli.py with reviewer.py acting as the orchestrator
• Reviews are also saved to pr_review_output.md