wasm-shell offers sandboxed agent shell

This is a very pragmatic alternative to handing an LLM a real shell or spinning up containers for every task. The real appeal is not just “WASM” but the tiny integration surface for building safer agent workflows.

• –Removes the usual docker/podman setup tax for agent tooling
• –Built-in commands plus custom programs make it useful out of the box for file ops and scripted workflows
• –Browser support makes it interesting for client-side sandboxes and demos, not just local dev
• –The virtual shell model narrows blast radius, though it still is not a full OS/container replacement for arbitrary binaries or deep isolation needs
• –The README hints at a good expansion path: MCP server support could make it even more useful for agent frameworks