Wraps agents/flow-orchestrator-designer/AGENT.md. Produces stages, steps, work-item assignees, interactive/background step mix, transition criteria, and restart/recall behavior.

Step 1 — Collect inputs

Ask the user:

1. Mode? design | audit

2. Target org alias (required — agent probes existing Orchestrations + referenced subflows)?

3. Workflow description (for design)?
   Example: "3-stage contract review with legal, procurement, and customer sign-off"

4. Assignees per stage? (user | group | queue | formula-driven)

5. Restart / recall requirements?

6. SLA per stage (if any)?

If the description has fewer than 2 stages, STOP — this is a single Flow, not an Orchestration.

Step 2 — Load the agent

Read agents/flow-orchestrator-designer/AGENT.md + mandatory reads (flow/orchestration-flows, flow/subflows-and-reusability, flow/screen-flows).

Step 3 — Execute the plan

Classify as Flow vs Orchestration (per standards/decision-trees/automation-selection.md).
Decompose into stages.
For each stage: steps, assignees, interactive/background mix, transitions.
Map each step to an existing subflow or a new one.
Emit restart/recall + cancellation semantics.

Step 4 — Deliver the output

Summary + confidence
Stage/step diagram (mermaid)
Per-step assignee + type
Subflow mapping
Restart/recall/cancellation behavior
Audit findings (inactive stages, orphaned subflows, missing fault paths)
Process Observations
Citations

Step 5 — Recommend follow-ups

/build-flow for each subflow that doesn't yet exist
/analyze-flow on the referenced subflows before deploying
/architect-perms if orchestration runs as different user contexts

What this command does NOT do

Does not deploy Orchestrations.
Does not build the screen flows / subflows referenced.
Does not replace BPM platforms — flags when the requirement outgrows Orchestration.