Building Sandboxes into OpenCode (Redirected — See Updated Articles)

This Article Has Been Split Into Two Focused Deep-Dives

The original Part 2 covered too much ground in a single article. It has been replaced by:

Part 2A: OS-Level Sandboxing — Kernel Isolation for AI Agents
Restrictiveness lattices, Bubblewrap, gVisor, Seatbelt, and the MCP server gap.

Part 2B: Application-Layer Defense — Stopping Exfiltration Inside the Sandbox
Input sanitization, SSRF defense, phantom credential proxying, content-addressed trust, WASM capability isolation, and the 9-gate threat matrix.

Part of the AI Agent Security series.