Cookbook
End-to-end, runnable recipes — each a full curl sequence from register to a resumed governed run, on a real vertical.
Each recipe runs a real vertical end-to-end: register → start a run → it parks at a
human gate → resume → outcome. Every step is a curl you can paste. The governance note
on each recipe calls out which step is irreversible (and therefore shadowed) and which gate
the run parks on.
Two shared conventions across every recipe:
gateon resume is the gate's step-id, not a display label. Start a run, readparked_at, and pass the step-id the run parked on. (GET /process-runs/{id}shows the full trace; the gate's step-id is the id in the run's events, e.g.bsa-gate,signoff.)- Irreversible tools stay shadowed (dry-run) unless a connector opted in with
reversible: true. During a governed rollout, leave them shadowed and let the gate collect a human verdict first — see Connect a step.
Setup for every recipe
Register once and export the token, then run the recipe's start/resume calls:
TOKEN=$(curl -s -X POST "$API/auth/register" \
-H "Content-Type: application/json" \
-d '{"email":"you@acme.com","password":"<your-password>","company_name":"Acme Ops"}' \
| jq -r .token)AML alert triage
Vertical aml — screens and monitors; parks at bsa-gate for the
BSA/AML Officer.
Prior authorization
Vertical prior-auth — parks at signoff for a medical director.
Debt collection
Vertical collections — parks at signoff for a manager.
Chargeback representment
Vertical chargeback-representment — parks at a fight-vs-refund gate.
Set up your team
Register (open), get your auto-created company, then invite members and assign roles — all company-scoped, so a teammate only ever sees your company's data.
AML alert triage
Onboard and screen a customer on the aml vertical, park at the BSA officer gate, sign it, and keep the run monitoring — every SAR and OFAC action stays shadowed until a human signs.