Unwind Documentation

Overview

This directory contains documentation for the Unwind reverse engineering skills library.

Unwind is a hybrid tool: a deterministic scanner (@unwind/core, tree-sitter) builds the verifiable ground truth (docs/unwind/.cache/scan-manifest.json) and LLM specialists add the semantic rebuild documentation. Completeness is verified by a deterministic manifest − docs diff rather than asserted.

Architecture (engine)

• packages/core — @unwind/core: file scan, language/category detection, rebuild-layer classifier, tree-sitter structural extraction (TS/JS, Python, Rust, Java, C#), import map, manifest builder, coverage diff.
• skills/scripts — bundled entry points the skills invoke: scan.mjs (→ scan-manifest.json), seed-layers.mjs (→ seeds/), and verify-coverage.mjs (→ coverage/ + gaps.md). _resolve-plugin-root.sh resolves the plugin root and lazily builds the core on first use.

The engine is an enhancement: if Node/pnpm/core is unavailable, the skills fall back to a pure-LLM flow.

The flow

start → scan.mjs → architecture.md
unwinding-codebase → seed-layers.mjs → specialists (seeded) → verify-coverage.mjs → completing → loop to 100%
synthesizing-findings → REBUILD-PLAN.md

Contents

• Main README - Project overview and installation
• Skills documentation - See individual skill directories under /skills

Guides

Coming soon:

• Getting Started Guide
• Skill Development Guide
• Best Practices for Reverse Engineering

Contributing

See the main README for contribution guidelines.