A local, file-based set of ten SKILL.md agent skills for task clarification, task tracking, pausing and handoff, profile memory, skill building, and writing.
ShadowSwaps/REA-Reconstruction-Companion
An open-source agent skill that helps AI agents understand existing software, plan a reconstruction, and validate it, working alongside the REA tool.
- Stars
- 131
- Forks
- 22
- Open issues
- 0
- Last commit
- 8 days ago
What is it?
REA-Reconstruction-Companion is a model-independent agent skill, written as Markdown instructions and reference guides in the Agent Skills format. It guides an AI agent through understanding how existing software behaves, turning those observations into requirements and a staged plan, and building a selected part only when explicitly requested. Each step keeps observed behavior, inferences, and design choices separate, and records gaps instead of assuming a reconstruction matches the original.
The skill is designed to pair with REA, which supplies the tools for examining software and collecting evidence. The companion helps the same agent use that evidence for requirements, design, implementation, and validation. An included Node.js 20+ installer supports Claude Code, Codex, and generic clients, with a dry-run preview, a verification check, and backups. Adding the optional REA integration inserts a marked rule into REA's skill file so relevant reconstruction requests also load the companion.
The project also sets explicit cost and permission limits: the agent is instructed not to start actions that could incur charges, and to skip any step whose no-charge status cannot be established. An optional, off-by-default update check runs at most once every seven days and still requires user approval before installing a new release.
Key features
- 01Turns observations, files, or recordings into behavior descriptions, requirements, and open questions.
- 02Produces specifications and staged reconstruction plans with evidence-linked acceptance criteria.
- 03Builds selected behavior in testable increments only when a build is requested.
- 04Includes validation guidance such as differential fuzzing, mutation tests, and interface contracts.
- 05Installer supports Claude Code, Codex, and generic skills directories, with dry-run and check modes.
- 06Optional REA integration adds a reversible routing rule to REA's skill file.
Use cases
- Understanding how an existing desktop or web app handles saving, editing, or import before planning a similar tool.
- Planning a reconstruction of one function or file format with documented inputs, outputs, and error cases.
- Rebuilding a selected workflow in a local project while reporting unverified differences.
- Recovering and changing an existing reconstruction while preserving compatibility with earlier work.
Quick start
- 1
Download or clone the repository from GitHub
- 2
Install the companion for Claude Code without changing REA
node install.mjs --client claude-code --integration off - 3
Preview changes before installing (optional)
node install.mjs --client claude-code --integration off --dry-run - 4
Verify the installation
node install.mjs --client claude-code --integration off --check
Things to know
- Requires Node.js 20 or newer to run the installer.
- Live investigation needs REA installed and configured separately for your client.
- Cannot be installed through chat-only or hosted upload clients.
In collections
Topics
FAQ
- What is REA-Reconstruction-Companion?
- An open-source agent skill that helps AI agents understand existing software, plan a reconstruction, and validate it, working alongside the REA tool. REA-Reconstruction-Companion is a model-independent agent skill, written as Markdown instructions and reference guides in the Agent Skills format. It guides an AI agent through understanding how existing software behaves, turning those observations into requirements and a staged plan, and building a selected part only when explicitly requested. Each step keeps observed behavior, inferences, and design choices separate, and records gaps instead of assuming a reconstruction matches the original.
- Is REA-Reconstruction-Companion free to use? What license does it have?
- REA-Reconstruction-Companion is open source under the MIT license. Check the license terms before using it in commercial products.
- Is REA-Reconstruction-Companion still maintained?
- Actively maintained. The last commit to REA-Reconstruction-Companion was on Oct 30, 2026.
- What can I use REA-Reconstruction-Companion for?
- Understanding how an existing desktop or web app handles saving, editing, or import before planning a similar tool. Planning a reconstruction of one function or file format with documented inputs, outputs, and error cases. Rebuilding a selected workflow in a local project while reporting unverified differences. Recovering and changing an existing reconstruction while preserving compatibility with earlier work.
- How do I install REA-Reconstruction-Companion?
- Download or clone the repository from GitHub Install the companion for Claude Code without changing REA: node install.mjs --client claude-code --integration off Preview changes before installing (optional): node install.mjs --client claude-code --integration off --dry-run Verify the installation: node install.mjs --client claude-code --integration off --check
- Where can I get REA-Reconstruction-Companion?
- The source code and installation instructions are on GitHub: https://github.com/ShadowSwaps/REA-Reconstruction-Companion
Summary generated by AI from the project's README. GitHub stats refresh daily.
Related projects
MCP server and CLI letting AI agents reverse engineer native binaries, Electron apps, .NET assemblies and websites.
A skills, routing and workflow layer for Hermes Agent that picks models per task lane and runs multi-step graph jobs in parallel.
A dev-only SDK and MCP server that lets AI coding agents drive your running web or desktop app and return a pass, fail, or cannot-tell verdict with evidence.
Configures AI coding agents like Claude Code, Cursor, OpenCode and Codex with persistent memory, shared workflows, skills and MCP servers.
A multi-platform hook and skill guard for AI coding agents that stops unrequested hashes, checksums, and scope creep in Codex, Claude Code, OpenCode and other hosts.