Supply chain orchestrator

Supply-Chain Orchestrator

The single entry skill for supply-chain work. It locates the task on the SCOR backbone (Plan → Source → Make → Deliver → Return) and delegates to one of 8 specialist spokes. The cross-cutting model every function shares — the service-level vs landed-cost trade-off under a binding constraint, plus the shared time/cost/quality conventions and escalation discipline — lives in supply-chain-core; read it before tuning safety stock, accepting a rate, or disposition-ing returns, because those are the same trade-off wearing different hats.

Cluster map (spoke → role)

• inventory-demand-planning — Plan: forecast demand, set safety stock, plan replenishment, model promo lift across SKUs/locations.
• production-scheduling — Make: sequence jobs, balance lines, optimize changeovers, resolve bottlenecks (TOC/SMED/OEE).
• quality-nonconformance — Make/quality gate: investigate NCRs, run root-cause, drive CAPA, interpret SPC, manage supplier quality.
• carrier-relationship-management — Deliver/source capacity: build the carrier portfolio, run RFPs, scorecard, allocate freight, negotiate rates.
• logistics-exception-management — Deliver/recover: handle delays, damage, loss, and carrier disputes; file freight claims; escalate.
• customs-trade-compliance — Deliver/cross-border: HS classification, Incoterms, duty optimization, restricted-party screening, penalty mitigation.
• returns-reverse-logistics — Return: RMA, receipt/inspection, disposition economics, refunds, return-fraud detection, warranty recovery.
• energy-procurement — Source (utilities): electricity/gas contracts, tariff & demand-charge optimization, PPA/hedging, multi-facility cost.

Routing rules (intent → spoke)

Plan & inventory

• "What should we forecast / how much to stock / when to reorder?" → inventory-demand-planning
• Promotion lift, seasonal transition, ABC/XYZ → inventory-demand-planning

Make (factory floor)

• Sequencing, line balance, changeover, bottleneck, disruption response → production-scheduling
• Defect, NCR, CAPA, SPC out-of-control, supplier quality escape → quality-nonconformance

Deliver (move it)

• Choose/negotiate/score carriers, RFP, routing guide, allocation → carrier-relationship-management
• Something went wrong in transit (late, damaged, lost, disputed) → logistics-exception-management
• Anything crossing a border (tariff, duty, Incoterm, sanctions screen, broker) → customs-trade-compliance

Return & utilities

• Product coming back, refund/disposition, warranty, return fraud → returns-reverse-logistics
• Energy contracts, tariffs, demand charges, PPAs → energy-procurement

Disambiguation when two could fire

• Late shipment because of customs hold → customs-trade-compliance for the clearance fix, logistics-exception-management for the customer/claim recovery.
• Stockout risk → inventory-demand-planning if it's a planning miss; production-scheduling if supply is constrained at the line; carrier-relationship-management if it's a capacity/transit gap.
• Returned unit that's defective → returns-reverse-logistics owns disposition; loop quality-nonconformance if it signals a systemic defect (CAPA trigger).

Standard flow

1. Locate the task on the SCOR backbone (Plan → Source → Make → Deliver → Return) and pick the function.
2. If it turns on a cost vs service trade-off, a buffer/safety level, a disposition decision, or an escalation threshold, pull the shared model from supply-chain-core first — these spokes optimize the same objective with different levers.
3. Delegate to the spoke(s). Multi-stage asks fan out in flow order (e.g. "launch a new SKU into a new country" → inventory-demand-planning → production-scheduling → carrier-relationship-management → customs-trade-compliance).
4. Return: chosen spoke(s), the service-vs-cost trade-off implied, the binding constraint, the buffer/threshold touched, and the next action.

Guardrails

See supply-chain-core. In short: protect the customer promise (OTIF / fill rate) at the lowest total landed cost — never trade service silently for unit-price savings, or vice versa. State the binding constraint before optimizing around it. Treat every buffer change (safety stock, lead-time padding, capacity reserve) as a cost-and-risk decision worth naming. Honor compliance hard stops (restricted-party hits, safety/quality holds, regulatory documentation) — these are non-negotiable gates, not trade-offs. Escalate on the thresholds each spoke defines rather than absorbing risk quietly.

Loading spokes on demand

To keep CLI startup context lean, this cluster's spokes are not separately registered as skills — only this orchestrator and its *-core are enumerated. When you route to a spoke named above, load it on demand by reading its file:

~/.agents/skill-clusters/skills/<spoke-name>/SKILL.md (or skills/<spoke-name>/SKILL.md inside the skill-clusters repo).