[connectors] Connector consent + write-scope safety + admin visibility

[serge-ivo]

[connectors] Connector consent + write-scope safety + admin visibility

Every connector today is read-only. The framework introduces writes (Sheets append, GitHub create-issue/PR, workflow trigger) operating with the user's real credentials — so consent + guardrails are a prerequisite, not an afterthought.

Do

• Declared scopes per connector (read vs write) surfaced to the subscriber at grant time: "this agent will be able to write to spreadsheet X / open PRs on repo Y."
• Consent record per (instance, connector, resource, scope); no write tool is offered without it.
• Approval gate on write/destructive tools by default (reuse the browser-agent human-approval pattern; ties to the trust-model issue [browser][trust] Trust & permission model for browser agents (design prerequisite) #75).
• Admin visibility: an admin page/panel listing every connector grant across users (provider, resource, scope, granted-at) — extends the admin portal transparency; audit every connector auth + write.

Acceptance

• A write tool is unavailable until the subscriber consents to that scope on that resource.
• Write/destructive calls require approval by default; all connector writes audited.
• Admin can see all connector grants + recent connector actions.

Part of the connector/tool framework epic. Gates the write features in the GitHub + Sheets connectors. Related: #75 (browser-agent trust model).

[serge-ivo]

Shipped — connector consent (lib/connector-consent.ts, instance_connector_consent, migration 0051); runRegistryTool enforces write-consent fail-closed before dispatch. Admin visibility via /v1/admin/connectors.