Reference Owner Agent Control Surface
Created Updated openspec/changes/complete-self-service-connection-onboarding/specs/reference-owner-agent-control-surface/spec.mdView on GitHub →
ADDED Requirements
Requirement: Owner-agent setup intents SHALL project the shared setup engine without exposing secrets
The owner-agent control surface SHALL initiate connector setup by projecting the shared owner-mediated setup engine. It SHALL return typed next steps, support states, deployment-readiness requirements, and proof-gate reasons, but SHALL NOT return provider credentials, owner-session credentials, browser session credentials, or grant-scoped MCP tokens.
Scenario: Owner agent initiates a supported setup flow
- WHEN a trusted owner agent initiates setup for a connector whose setup path is supported in the current deployment context
- THEN the response SHALL include a typed next-step kind from the setup engine
- AND the response SHALL identify whether the owner must open a URL, enroll a collector, capture a credential through an owner surface, upload a file, or complete another owner-mediated action
- AND it SHALL NOT include any provider secret or owner-session bearer
Scenario: Owner agent initiates a proof-gated connector
- WHEN a trusted owner agent initiates setup for a connector whose setup path exists only behind a live-proof gate
- THEN the response SHALL be typed as proof-gated or unsupported with a concise reason and documentation pointer
- AND it SHALL NOT advertise a connection as active or supported before the proof gate is closed
Scenario: Owner bearer is presented to MCP for setup
- WHEN a caller presents an owner-agent bearer to
/mcpin order to add or manage connections - THEN the reference SHALL reject the bearer for MCP tool access
- AND it SHALL point the caller toward owner-agent REST, console, or CLI setup surfaces as appropriate
Requirement: Owner-agent setup SHALL distinguish connector templates, setup plans, drafts, and active connections
The owner-agent control surface SHALL distinguish connector templates from setup plans, setup plans from draft or pending setup state, and draft/pending setup state from active connections. It SHALL NOT report a connector as an active connection merely because a setup intent was created.
Scenario: Intent returns an owner next step
- WHEN a trusted owner agent creates a setup intent and the owner has not yet completed the required owner-mediated step
- THEN the response SHALL identify the setup intent or next step as pending setup
- AND connection read surfaces SHALL NOT show a new active connection unless the connector instance lifecycle has crossed its activation proof boundary
Scenario: Setup fails before activation
- WHEN setup fails before the activation proof boundary
- THEN owner-agent status SHALL return a typed failed or blocked setup state with non-secret remediation guidance
- AND it SHALL NOT leave behind a visible zero-record active connection as a false success