7 · Connect services

Use case · the want as a person would post it:

“Wire my form signups into my mailing list and my CRM — whatever works.”

PARTIAL The purest test of the provider lane: nothing here needs money or outreach, only the ability for a person to connect accounts and for an agent to hold scoped capabilities against them. The broker and generic MCP execution lane now exist. A third-party automation provider, production connection, and passing end-to-end action remain unproven.

The walk, step by step

Deal signed
FINISH

Blocks —

Gates that fire Signing freeze.

Map the flow: sources, destinations, fields
PROVIDE

Blocks structured_form, written_response

Gates that fire Works.

Connect the automation platform (or each service)
GRANT

Blocks connect_account, grant_access

Gates that fire REJ-11 what/why/scope/until; equivalent-swap lets the agent substitute a same-shape service without re-asking.

Note FOUNDATION BUILT / PROVIDER UNPROVEN: scoped authorizations, exact-action approval, execution, and receipts exist; platform-specific connection and execution still need proof.

Build + test the zap; show a passing test run
APPROVE

Blocks review_approve

Gates that fire Works.

Finish line
FINISH

Blocks yes_no

Gates that fire Binary only.

Gaps & what to build

Equivalent-swap is unusually valuable here: “substantially the same capability under the same limits” is exactly the choice between one automation platform, another, or a native webhook. The walk and generic connector lane now carry the law and execution shape. What remains is provider coverage, production configuration, and a passing receipt-backed run.

HAR Reference · updated 2026-08-30 from target_walk.py, connector_broker.py, connector_registry.py, and connector_executors.py · staging is source of truth