Anatomy of a workflow - Frayme

Every workflow is a JSON object with five top-level concerns: metadata, parameters, nodes, edges, and the run context that those nodes share at execution time.

Top-level shape

{
  "id": "wf-...",
  "name": "...",
  "description": "...",
  "status": "draft" | "active" | "archived",
  "version": 2,
  "createdAt": "...",
  "updatedAt": "...",
  "createdBy": "...",
  "parameters": [ ... ],
  "nodes": [ ... ],
  "edges": [ ... ]
}

See Variables & templating for how parameters and node outputs are referenced. See Node reference for the per-node data schemas.

Terminal output nodes

Every path through a workflow ends in an output node. The output carries:

a label (APPROVED, REJECTED, IN_REVIEW, REPORTED_TO_REGULATOR, …) — surfaces as the terminal state in the audit log.
a schema — explicit list of fields written into the final decision record.
an optional returnAllFields boolean — if true, the entire run context is persisted with the decision; otherwise only the declared schema fields.

{
  "id": "out-approved",
  "type": "output",
  "data": {
    "label": "APPROVED",
    "returnAllFields": false,
    "schema": [
      { "name": "case_id",          "type": "string",  "required": true },
      { "name": "decision",         "type": "string",  "required": true },
      { "name": "risk_score",       "type": "integer", "required": true },
      { "name": "risk_band",        "type": "string",  "required": true }
    ]
  }
}

A workflow can have multiple terminal outputs — one per resolution class. Common shapes:

Workflow type	Typical terminals
Onboarding (KYC / KYB)	`APPROVED`, `IN_REVIEW`, `REJECTED`
Transaction monitoring	`APPROVED`, `BLOCKED`, `REPORTED_TO_REGULATOR`
Lifecycle webhook consumer	`case_updated`
Sub-flow	Single named outcome (`link_delivered`, `case_closed`, …)

returnAllFields: true is used for sub-flows that pass their entire context downstream and for audit-heavy terminals (KYB approvals, regulator submissions) that benefit from the full evidence pack on the decision. Otherwise prefer an explicit schema — the decision record stays compact and the audit export bundles read better.