Capability is powerful.
Authority is specific.

Configuration is not authority.

A skill describes what an agent can do. Execution authorization decides whether a specific action may proceed. Assigning a skill does not silently grant access to every system it mentions.

Connections are not permission.

Installation, enablement, credentials, permissions, policy decisions, approval, and execution are separate boundaries. A connector must not turn broad credentials into ambient agent authority.

Imported content remains untrusted.

Discovery should inspect files without running imported scripts. Model classifications are suggestions with evidence and review state. They are not an authorization source.

Decisions need evidence.

The LNSAT direction separates proposed actions, policy evaluation, approvals, authorization, connector invocations, receipts, and audit evidence. Additional engines must preserve explicit execution boundaries.

Report responsibly.

Do not publish credentials, customer data, or exploit details in a public issue. Use the repository’s private vulnerability reporting feature if enabled, or establish a maintainer-approved private channel. See SECURITY.md in the source repository.

Scope of this website.

The public website is a static product introduction and design preview. It does not run agents, accept account credentials, provide a hosted command center, or claim security certification.