Static copy

Copied from docs/en/governance/trust-model.md in https://github.com/poppy-team/prumo (MIT). Pinned to revision e91694d3959be1ed92757b34063efe0b2191821f, blob eb04c40165ed03fd933415a065a008c9927fb3ff. The source repository remains canonical; this copy is refreshed through a sync pull request, not live.

Trust & Security Model

Prumo operates under a model of least privilege and strict isolation. Autonomous agents receive only the permissions explicitly granted by the project profile.

Trust Levels

1. Canonical Repository (Highest Level): Files versioned in Git with intact hashes. No AI can mutate these files without passing gate validation.
2. Local Development Environment: The human user holds supreme authority and can revoke or abort any harness execution at any time.
3. AI Agents (Restricted Level): Run in directive sandboxes, with no access to secrets or production environment variables, and with commands restricted to the profile's allowlist.
4. External Sources & Web (Untrusted): Any data retrieved from the internet or from third-party packages is considered untrusted until explicitly validated for compliance.