Agent Skills: Product Orchestrator

Product Orchestrator

Use When

Use while an idea still needs bounded product definition before architecture or implementation.

Route Away

Use moonshot-architecture after product approval and moonshot-orchestrator only for bounded implementation-ready work.

Role

Run the product-definition workflow before code-oriented Moonshot execution.

This is the default public entrypoint for the Intake stage when the request is still shaping product scope.

This skill is explicitly for:

• idea to product intent
• product intent to PRD
• PRD to product behavior model
• behavior model to architecture
• architecture to execution slices

This skill is not for:

• market validation
• user interview automation
• MVP experiment pipelines
• shipping code directly

It may prepare a demo_first MVP execution pack. That pack is a planning and execution contract, not a market experiment runner.

Output Contract

Write artifacts under:

• {tasksRoot}/{feature-name}/product/

Required outputs:

• PRODUCT_INTENT.md
• PRD.md
• SOLUTION.md
• SPEC.md
• ADR/*.md
• PLAN.md
• tasks/*.md
• ASSUMPTIONS.md
• BLOCKERS.md
• execution/REQUIREMENTS_TRACEABILITY.md when document-trace completion is required
• execution/SCENARIO_MATRIX.md when user-visible flows matter
• execution/UAT_CHECKLIST.md when the target is UAT-ready handoff

Conditional demo-first MVP outputs:

• MVP_SCOPE.md
• MINI_ARCHITECTURE.md
• UI_DEMO_PLAN.md
• UI_FLOW_MAP.md
• UI_STATE_MATRIX.md
• MOCK_SCENARIOS.md
• MOCK_API_CONTRACT.md
• USER_DEMO_TEST.md
• DEMO_EVIDENCE.md
• USER_DEMO_APPROVAL.md
• POST_DEMO_IMPLEMENTATION_PLAN.md
• UI_CHANGE_REQUEST.md

Planning artifacts should also record:

• explicit non-goals
• scope reduction or scope hold notes when requests are too large
• a short cost/benefit rationale at PRODUCT_INTENT, PRD, and PLAN
• canonical domain terms or glossary gaps when language is ambiguous
• testing decisions focused on user-visible behavior, not implementation details

Procedure

1. Build compact intake/plan knowledge context and classify facts, decisions, assumptions, and blockers. 1.1. classify unresolved input as fact, decision, assumption, or blocker; do not self-resolve decisions affecting scope, security, data, package/runtime surface, or user-visible behavior.
2. Draft PRODUCT_INTENT.md, PRD.md, SOLUTION.md, SPEC.md, and PLAN.md in order.
3. Run the product gate at every stage; add CEO review for intent, PRD, and plan, plus engineering review for spec and plan.
4. Use Discovery Map and current-fact references only when needed; they remain advisory and never authorize execution. 4.1. Record task-relevant consultation through docs/public/guidelines/skill-readiness-policy.md.
5. Slice accepted planning into tasks, retry a weak draft at most twice, and reduce scope when value is unclear.
6. Hand the approved package and compact context to the appropriate execution entrypoint. 6.1. Route architecture-heavy PRDs through moonshot-architecture, then preserve REQUIREMENT_INVENTORY.md, TRACEABILITY_MATRIX.md, and ARCHITECTURE_REVIEW.md for moonshot-orchestrator or moonshot-phase-runner handoff.

Hard Stops

Every stage ends with one of:

• pass: ready for the next stage
• conditional_pass: acceptable with explicit assumptions or follow-up notes
• fail: rewrite the current stage

Escalation rules:

• Same issue repeated twice -> conditional_pass
• Missing but non-critical detail -> add to ASSUMPTIONS.md
• Hard dependency missing -> add to BLOCKERS.md
• Weak value or poor cost/benefit -> reduce scope, hold scope, or fail the stage

Stage-specific detail, value tests, and demo-first sequencing are conditional references in docs/public/guidelines/product-definition-workflow.md and docs/public/guidelines/demo-first-mvp-gate.md.

Approval Boundary

• Human approval may be used to accept the planning package before execution begins.
• After execution begins, implementation -> review -> verify -> retry loops should continue without additional human checkpoints unless a true blocker or external dependency appears.
• Exception: demo_first MVP work must hard-stop after Mock Functional Demo evidence until USER_DEMO_APPROVAL.md is approved with non-empty approved scope.

Handoff Contract

When the plan passes:

• provide document paths, not full inline content
• summarize assumptions and blockers
• include architecture package paths when moonshot-architecture was used: REQUIREMENT_INVENTORY.md, ASR_CATALOG.md, TRACEABILITY_MATRIX.md, selected ADR/*.md, and ARCHITECTURE_REVIEW.md
• hand tasks/*.md to the implementation-oriented workflow
• route bounded implementation to moonshot-orchestrator; route multi-phase, staged adoption, or long-running packages to moonshot-phase-runner

Recommended next step:

• /moonshot-orchestrator with the generated product package

References

• docs/public/guidelines/product-definition-workflow.md
• docs/public/guidelines/demo-first-mvp-gate.md
• docs/public/guidelines/retrieval-and-recency-policy.md
• templates/product-definition/DISCOVERY_MAP.template.md
• templates/product-definition/DISCOVERY_TICKET.template.md
• schemas/discovery-map.schema.json
• docs/public/guidelines/skill-readiness-policy.md
• <MOONSHOT_RELAY_HOME>/templates/product-definition/
• skills/product-gate-reviewer/SKILL.md
• skills/plan-ceo-review/SKILL.md
• skills/plan-eng-review/SKILL.md
• skills/task-slicer/SKILL.md
• skills/assumption-ledger/SKILL.md
• skills/moonshot-orchestrator/SKILL.md

Project Knowledge Context Contract

Product-definition work uses advisory projectKnowledgeContext with stage=intake or stage=plan before plan-package prompt assembly. The context is a compact recall source, not an enforcement source.

If the helper is unavailable, continue with degraded advisory metadata unless the user explicitly requested a strict memory task. Do not inline raw MemoryGraph/KG/ontology records, logs, transcripts, or secrets into product prompts or plan artifacts.