RFC-0017: Security and Capability Model

Status: Draft | Version: 0.1.0

1. Principle

Authority is explicit. Possession of a pointer or handle does not imply permission.

2. Capabilities

READ_WORLD, WRITE_COMPONENT, CREATE_ENTITY, DESTROY_ENTITY, EXECUTE_SYSTEM, GPU_ACCESS, NETWORK_SEND, NETWORK_RECEIVE, RESOURCE_READ, RESOURCE_WRITE, JIT_INSTALL.

3. Scope

Capabilities are scoped to World, Region, Entity set, Component type and lifetime.

4. Validation

Every privileged operation checks capability before effect.

5. Plugin isolation

Untrusted plugins should run out-of-process or in a sandboxed execution environment. In-process plugins are trusted code.

6. Generated code

JIT/AOT artifacts carry capability manifests. The runtime refuses installation if required capabilities exceed policy.

7. Audit

Security-sensitive mutations produce audit events containing actor/module identity, capability, WorldVersion and operation.

8. Acceptance

No component mutation, ownership transfer, resource write or network side effect can occur without an explicit authority path.