Decision-time controls
The planned model evaluates consequential requests with their target, scope, policy, and expected side effect.
Security
Make authority explicit.
Make evidence useful.
Design security-agent operations around declared capabilities, bounded connector actions, policy context, specific approvals, and reconstructable execution evidence.
Product direction · In active development
01
Configuration is not authority
Giving an agent a security skill should not silently authorize every operation that skill describes.
02
Review the actual request
Approval designs should name the requesting agent, resource, data, risk, policy, scope, expiration, and supporting evidence.
No vague grants
The objective is to approve or deny an actual operation rather than an indefinite permission.
03
Reconstruct the path
Audit Explorer direction links proposal, effective configuration, policies, approvals, authorization, connector result, receipt, and evidence.
Preview status
Security workflows and integrations are in active development.
