Skip to content

Installation

Requirements

  • Node.js 18+ installed on your machine

If you don't have Node.js, install it at nodejs.org and come back here.

One command, that's all

In the root of the legacy project you want to analyze:

npx reversa install

The installer does all of this for you:

  1. Detects the AI engines present in the environment (Claude Code, Codex, Cursor, Gemini CLI, Windsurf)
  2. Asks which agents to install (all selected by default)
  3. Collects project name, language, and preferences
  4. Copies agents to .agents/skills/ and .claude/skills/ (for Claude Code)
  5. Creates the engine entry file (CLAUDE.md, AGENTS.md, etc.)
  6. Creates the .reversa/ structure with state, configuration, and plan
  7. Generates the SHA-256 manifest for safe future updates

It's like npm install, but for your reverse engineering agent team.

What gets created in the project

legacy-project/
├── .reversa/               ← analysis state, config, and context
├── .agents/skills/         ← universal agents (all engines)
├── .claude/skills/         ← mirror for Claude Code
├── CLAUDE.md               ← entry point for Claude Code (if detected)
├── AGENTS.md               ← entry point for Codex (if detected)
└── _reversa_sdd/           ← where specs will be generated (empty initially)

Your files stay intact

The installer only creates new files. It never modifies or deletes any existing file in your project.

Backup before starting

Strong recommendation: make a backup

Although Reversa never modifies your files, AI agents can make mistakes. Before starting the analysis:

  1. Make sure all files are committed in Git
  2. Have the repository on GitHub, GitLab, or Bitbucket
  3. Make a local copy of the folder as extra safety: cp -r my-project my-project-backup

If something unexpected happens, git restore . fixes it.

Adding another engine later

If you want to add support for another engine later (for example, you installed only for Claude Code and now want Codex too):

npx reversa add-engine

The installer detects what already exists and adds only what's missing.