Scope Syntax

Scope is expressed as an array of service permission objects. Each object names a service and the actions the agent requires within that service.

Format

{
  "service": "<service-slug>",
  "actions": ["<resource>:<action>", "..."]
}

service, resource, and action are lowercase ASCII identifiers containing letters, digits, and hyphens, and must start with a letter. An action is written as exactly one resource:action pair. A service object must not repeat an action, and a declaration must not repeat a service.

Identity providers define the action vocabulary for each service. * is allowed only as the action segment (resource:*) and represents every action the provider recognises for that resource.

Common action names

Action	Meaning
read	Read an individual resource
list	List or search a resource collection
create	Create a resource
update	Update a resource
write	Provider-defined create and/or update access
delete	Delete the resource
*	All provider-recognised actions for the resource

Examples

{ "service": "stripe", "actions": ["payments:read", "customers:read"] }
{ "service": "github", "actions": ["repos:read", "issues:write"] }
{ "service": "cloudflare", "actions": ["dns:*"] }
{ "service": "vercel", "actions": ["env:write"] }

Principle of least privilege

Declared scope should reflect the minimum the agent requires for its stated task. Implementations may flag agents whose declared scope is significantly broader than their observed usage.

For policy narrowing, an exact declared action may only be granted exactly. A declared wildcard such as dns:* may be narrowed to one or more explicit actions for dns, but no other resource or service.