sandeco/reversa

Transform legacy systems into executable specifications for AI coding agents

What it solves

Reversa solves the problem of "trapped knowledge" in legacy software systems. When production systems accumulate years of undocumented business rules and architectural decisions, AI coding agents often cannot safely evolve them because they lack a formal specification of what the system is actually doing. Reversa bridges this gap by reverse-engineering existing code into operational specifications that AI agents can use as contracts to evolve the system without breaking existing logic.

How it works

Reversa is installed as a framework within a legacy project. It coordinates a team of specialized AI agents (such as the Scout, Archaeologist, and Detective) through a multi-phase "Discovery" pipeline (Reconnaissance, Excavation, Interpretation, Generation, and Review). These agents analyze the codebase, extract business rules and module contracts, and generate traceable specifications (SDD specs) in a dedicated output folder.

Once these specifications exist, the framework provides specialized workflows for different goals:

  • Evolution: Moving from specs to new code via the /reversa-forward pipeline.
  • Migration: Planning a rebuild on a modern stack using /reversa-migrate.
  • Documentation: Rendering the extracted knowledge into an HTML mini-site via /reversa-docs.
  • Greenfield: Starting a new project from a raw idea using /reversa-new.

Who it’s for

Software engineers and architects working with legacy systems who want to use AI agents to maintain, evolve, or migrate their software without the risk of introducing regressions due to missing documentation.

Highlights

  • Non-destructive analysis: The framework guarantees that it never modifies or deletes existing project files; all writes are restricted to specific .reversa/ and _reversa_sdd/ folders.
  • Agentic Orchestration: Uses a team of 10 specialized agent teams (Discovery, Ideation, Migration, etc.) to handle different stages of the software lifecycle.
  • Operational Contracts: Produces specifications designed for AI agents rather than just human-readable documentation.
  • Agent Agnostic: Does not store API keys; it delegates intelligence to the AI agent already present in the user's environment (e.g., Claude Code, Cursor, Codex).
  • Causal Traceability: Includes dedicated bug agents to track and fix defects with direct links back to the specifications.

Related

  • Project
    kalil0321/reverse-api-engineer

    A tool that turns websites into APIs by capturing network traffic via a browser or AI agent and using an LLM to generate a typed API client in multiple languages.

  • Project
    angusdevgo/Seep-Reverse-Lab

    Seep Reverse Lab is an AI‑driven, multi‑platform reverse‑engineering workbench. You speak plain‑language goals (e.g., “bypass the premium check in FooBar.apk”), and the built‑in agent automatically identifies the binary’s platform, routes the job through a unified MCP server that wraps 23 tools (Radare2, JADX, Apktool, Frida, IDA Pro, etc.), runs the analysis, generates a self‑healing PoC, confirms the result in “airplane‑mode” (CWE‑602 audit), and finally produces a three‑part security report. The system includes a knowledge base of 289 journals, token‑budget‑aware output modes, and one‑click Windows/Linux installation.

  • Project
    WhiteNightShadow/firefox-reverse

    Firefox-Reverse is a specialized browser that integrates an AI agent and isolated fingerprint management to automate the reverse-engineering of complex web security parameters like signatures and tokens.

  • Project
    makecindy/cindy

    Cindy is an open‑source, cross‑platform AI‑agent client (Electron desktop + React‑Native mobile) that integrates multiple LLM harnesses (Claude Code, Codex, etc.) with local file access, browser/OS control, persistent memory, and extensible “skills”. The repo contains the client code, shared TypeScript packages, and tooling to build/run the apps; the backend service lives elsewhere. It supports both cloud‑backed usage (via a Cindy account) and fully local agents (skip sign‑in). Licensed under Apache‑2.0.

  • Project
    OthmanAdi/planning-with-files

    A persistent file-based planning system for AI coding agents that stores task plans, findings, and progress on disk to prevent state loss during context resets.