Vouch request: wearethecompute — inter-agent messaging for sandboxes

[wearethecompute]

What do you want to work on?

I maintain OpenFused — a decentralized context mesh for AI agents. It provides encrypted agent-to-agent messaging via plain files: Ed25519 signed, age encrypted (X25519 + ChaCha20-Poly1305), with a GPG-style keyring and a public agent registry.

OpenShell provides the sandboxed walls. OpenFused provides the mailboxes between rooms.

Proposed contribution

An integration that lets OpenShell sandboxes communicate with each other securely:

1. Mount openfuse stores as sandbox volumes — each sandbox gets its own context store (inbox/, shared/, PROFILE.md)
2. Policy-aware messaging — the OpenShell policy engine controls which peers a sandbox can communicate with (filesystem policy restricts inbox writes to approved agents)
3. Zero-config agent discovery — agents register on the public registry, sandboxes discover and message each other by name
4. Encrypted by default — messages between sandboxes are age-encrypted end-to-end, the host/gateway never sees plaintext

This could be:

• A docs page showing how to use OpenFused with OpenShell today (mount a store volume, configure peers)
• A sandbox policy template for inter-agent messaging
• An agent skill that wraps the openfuse CLI for sandbox-to-sandbox communication

Why this matters

OpenShell sandboxes are currently isolated by design — no built-in way for agents in different sandboxes to collaborate. File-based messaging fits naturally: agents already read/write files, the policy engine already controls filesystem access, and signed+encrypted messages maintain the security model.

About the project

• npm: npm install -g openfused (v0.3.2)
• crates.io: cargo install openfuse
• Docker: docker compose up
• Registry: https://openfuse-registry.wzmcghee.workers.dev
• MCP server: 13 tools for Claude Desktop/Code/Cursor
• MIT licensed

https://github.com/wearethecompute/openfused