Security
VaultCrux Shield
Security controls are presented as operational behavior, not slogans.
Shield modules
- Trust registry: pin and validate trusted server digests.
- Capability firewall: deny by default; explicitly allow safe capability paths.
- Taint engine: tainted content cannot trigger dangerous actions.
- Approval gate: propose, approve, execute with signed receipts.
- Sandbox runner: isolate risky tools during execution.
- UI sandbox: restrict untrusted tool UI behavior.
Tool UI cannot present approvals; only VaultCrux chrome can.
In plain English: if content is tainted or untrusted, it can be read, but it cannot directly trigger dangerous actions.