← Principles
validated

“Make trust scopes visible and editable.”

Trust granted invisibly is trust that cannot be revoked deliberately. When what an agent may do — and what it remembers — is a first-class object the human can inspect and edit, authorization becomes a design surface instead of an accumulating liability.

The Context Matrix

Where this principle has been tested. Untested cells are honest gaps, not passing grades — seen evidence for one? Tell us.

ChatIDECLICanvas / DocAmbient / Background
Coding ● ◐
Knowledge Work ●
Consumer ●
Enterprise Ops
Creative

● supports · ◐ boundary · ✕ violates · blank untested

Linked Patterns

ApprovalDelegationMemoryGenerative UI

Boundary Conditions

  • coding × ambient-background × background

    Mid-run scope editing is unreachable for background agents — the scope you granted at dispatch is the scope you live with until the run ends. Visibility holds; editability breaks.

Evidence

supports

Claude Code allow rules: granted permissions persist as an inspectable, editable list.

coding × CLI × supervised · Claude Code: permission modes and plan approval

boundary

Mid-run scope editing is unreachable for background agents — the scope you granted at dispatch is the scope you live with until the run ends. Visibility holds; editability breaks.

coding × ambient-background × background · Codex: cloud tasks as delegation unit

supports

ChatGPT memory: review-and-delete settings surface makes what the agent retains editable.

consumer × chat × supervised · ChatGPT: memory with review and delete controls

supports

MCP Apps: declared surface contracts make what a third-party server may render an explicit, inspectable scope.

knowledge-work × canvas-doc × auto-with-gates · MCP Apps: interactive UI from servers

Tagged Stories