Confining agents
The envelope decides what stops for a human. These decide what is reachable at all.
- Why contenox confines agents — the threat model this answers
- Scoping what an agent may do — narrowing a toolset before anything runs
- The sandbox wall — Landlock and namespaces around a foreign agent
- Trusted binaries — pinning what a shell command is allowed to be
- Least-privilege shell environment — what a spawned command inherits
- External ACP agents — register and verify a foreign agent inside the wall