unravel-review 0.2.0

AI-powered CLI that decomposes PR diffs into causal threads for human code reviewers

Unravel

AI-powered CLI that decomposes PR diffs into causal threads for human code reviewers.

Instead of reviewing a diff file-by-file, Unravel groups changes by why they were made — each thread tells a story from root cause to effect, so reviewers build understanding progressively.

Installation

pip install unravel-review

Quick Start

# Analyze the last commit
unravel diff HEAD~1..HEAD

# Analyze a range
unravel diff main..feature-branch

# Analyze a GitHub PR (requires gh CLI)
unravel pr 42

# JSON output for piping
unravel diff HEAD~1..HEAD --json | jq .

# Compact tree view
unravel diff HEAD~1..HEAD --tree-only

Requirements

• Python 3.12+
• Git
• An API key for a supported LLM provider
• GitHub CLI (only for unravel pr)

Configuration

Set your API key as an environment variable:

export ANTHROPIC_API_KEY=sk-ant-...

CLI Options

Flag	Description
--model, -m	Model to use (default: claude-sonnet-4-6)
--provider, -p	LLM provider (default: anthropic)
--json, -j	Output raw JSON
--tree-only, -t	Compact tree view
--thinking-budget	Extended thinking token budget (default: 10000)
--staged	Analyze only staged changes (diff command only)
--remote	Git remote name (pr command only, default: origin)

Environment Variables

Variable	Description
ANTHROPIC_API_KEY	API key for Anthropic/Claude
OPENAI_API_KEY	API key for OpenAI (future)
UNRAVEL_PROVIDER	Default provider
UNRAVEL_MODEL	Default model
UNRAVEL_THINKING_BUDGET	Default thinking budget

How It Works

1. Extract — Pulls the diff from git or GitHub
2. Parse — Breaks the diff into hunks using unidiff
3. Analyze — Sends hunks to an LLM with a specialized prompt that decomposes changes into causal threads
4. Validate — Checks all hunks are covered and thread dependencies are consistent
5. Render — Displays threads in rich terminal output, JSON, or tree view

Multi-Provider Support

Unravel is designed to work with multiple LLM providers. Currently supported:

• Anthropic (Claude) — default, with extended thinking support

Planned: OpenAI, Gemini, and more. The provider abstraction is ready — contributions welcome.

Development

git clone https://github.com/roo-oliv/unravel.git
cd unravel
uv venv --python 3.12 .venv
source .venv/bin/activate
uv pip install -e ".[dev]"

# Run tests
pytest

# Lint
ruff check src/ tests/

License

AGPL-3.0-or-later