OmniWire

Infrastructure layer for AI agent swarms — 88 MCP tools for mesh control, A2A protocol, OmniMesh VPN, CyberSync, web scraping, firewall management, browser a...

Do not let the agent automatically run the 'npm install -g' command or supply private SSH keys without review. Before installing: (1) inspect the omniwire npm package contents or the GitHub source and release signatures; (2) prefer installing manually under your control instead of letting the agent exec installation; (3) do not reuse personal SSH keys — create limited, purpose-built deploy keys on each target host and reference those; (4) confirm which native tools (aria2c, netcat, lz4) are required and install them deliberately; (5) verify how CyberSync/Postgres credentials are provided and where network ports will be exposed; (6) consider running the tool inside an isolated environment (VM/container) first and remove ~/.omniwire and any keys if you uninstall. If you want, share the npm package tarball or the GitHub release contents and I can point out specific lines that need special attention.

Type: OpenClaw Skill Name: omniwire Version: 3.5.0 The 'omniwire' skill provides an exceptionally broad and high-risk feature set, including remote SSH execution across a mesh, VPN management, and browser automation. Most concerning are the 'CyberSync' tools, which specifically target and sync sensitive configuration and memory files from other AI agents (e.g., Claude Code, OpenClaw), and tools for extracting browser cookies (`omniwire_cookies`) and managing TOTP codes (`omniwire_2fa`). While these capabilities are framed as infrastructure management, the concentration of access to credentials, secrets, and remote systems across multiple nodes creates a massive attack surface and potential for abuse.