Oracle Solution Design Orchestrator

Replaces: claude-flow swarm orchestration for solution design workflows
Method: Native Claude Code Task agents with parallel dispatch + quality gates
Confidentiality: Conversation-only context model. Zero persistent state.

When to Use This Skill

Invoke when the user needs:

• A complete solution design for a customer engagement
• A Solution Design Document (SDD) for an OCI-based solution
• An architecture recommendation with BOM and visuals
• A prototype or interactive demo for a customer presentation

Trigger: /oracle-solution-design or when user says "solution design", "SDD", "architecture for [client]"

---

The 5-Phase Workflow

  DISCOVER --> ARCHITECT --> VISUALIZE --> PROTOTYPE --> DELIVER
     |             |             |              |            |
  [Gate 1]     [Gate 2]     [Gate 3]       [Gate 4]    [Gate 5]
     |             |             |              |            |
  Parallel     system-       nano-banana    coder         Package
  research     architect     + infogenius   agent         + audit
  agents       agent

Phase Execution Model

Each phase uses native Claude Code capabilities. No daemon, no vector DB, no external processes.

Phase	Agent Strategy	Tools Used
DISCOVER	Parallel Task(researcher) x2-3	WebSearch, WebFetch, Grep, Read
ARCHITECT	Task(system-architect) + Task(researcher)	Read, WebFetch, Write
VISUALIZE	Direct (main context)	Nano Banana MCP, Read
PROTOTYPE	Task(coder) or direct	Write, Edit, Bash
DELIVER	Direct (main context)	Write, Read, Bash

---

Phase 1: DISCOVER

Objective

Gather all information needed to design the solution. Run research in parallel.

Parallel Agent Dispatch Pattern

Launch 2-3 Task agents simultaneously:

Agent 1 -- OCI Capabilities Research:

Task(subagent_type="researcher")
"Research OCI services for [use case]. Check:
 1. OCI GenAI models available (docs.oracle.com/iaas/Content/generative-ai/)
 2. Relevant oracle-quickstart repos (github.com/oracle-quickstart)
 3. OCI AI Blueprints if GenAI workload
 Return: service list with capabilities and limitations."

Agent 2 -- Industry/Domain Research:

Task(subagent_type="researcher")
"Research [industry/domain] AI patterns:
 1. Common architecture patterns for [use case]
 2. Compliance/regulatory requirements
 3. Industry benchmarks and KPIs
 Return: requirements matrix and constraints."

Agent 3 -- Pricing and Competitor Landscape (if needed):

Task(subagent_type="researcher")
"Research current pricing for [specific OCI services]:
 1. Verify at oracle.com/cloud/price-list/
 2. Check OCI Cost Estimator for complex scenarios
 3. Note any commitment discounts or free tiers
 Return: verified pricing with sources and dates."

Confidentiality Gate

Before proceeding, verify:

• No real customer names in any output
• All context uses codename only
• Research results are generic (industry patterns, not client-specific)

Output

Mental model of the solution. No document yet. Just validated understanding.

Quality Gate 1: Discovery Complete

• Core use cases identified (top 3)
• OCI services mapped to requirements
• Constraints documented (compliance, timeline, budget)
• Existing OCI patterns checked (oracle-quickstart, AI Blueprints)

---

Phase 2: ARCHITECT

Objective

Design the technical architecture with OCI service selection and cost estimation.

Agent Strategy

Primary: System Architect Agent

Task(subagent_type="system-architect")
"Design an OCI architecture for [use case] with these requirements:
 [paste discovery results]

 Deliver:
 1. OCI service selection with rationale for each choice
 2. Architecture tiers (Basic / Advanced / Premium)
 3. Data flow: ingestion > processing > storage > retrieval > presentation
 4. Security architecture: IAM, encryption, network isolation
 5. Integration points and APIs"

Support: Pricing Verification

Task(subagent_type="researcher")
"Verify pricing for these OCI services: [list from architect]
 Source: oracle.com/cloud/price-list/
 Return: Monthly cost estimate by tier with line items."

MANDATORY Checks Before Finalizing Architecture

1. AI Blueprints decision tree -- Does an existing blueprint cover this? (See CLAUDE.md)
2. Model selection -- Use the OCI GenAI Model Selection Matrix (See CLAUDE.md)
3. Pricing claims -- NEVER say "X times cheaper" without specific model comparison
4. Service availability -- Confirm services exist in target OCI region

Document Generation

Generate the SDD in clients/[CODE]/ using /oracle-sdd-generator skill.

Quality Gate 2: Architecture Validated

• Every OCI service verified against official docs
• Pricing confirmed via oracle.com/cloud/price-list/ (not from memory)
• No blanket cost comparison claims
• Existing patterns checked (oracle-quickstart, technology-engineering)
• Architecture follows OCI Well-Architected Framework principles
• Confidentiality audit passed (no client names, codename only)

---

Phase 3: VISUALIZE

Objective

Create architecture diagrams that are logo-free and technically accurate.

Direct Execution (Main Context)

Visual generation stays in the main context for interactive refinement:

1. Invoke /oracle-infogenius or Nano Banana directly
2. Mandatory prompt elements:
   • "NO Oracle logos -- text labels only"
   • Oracle Red (#C74634) for accents
   • Dark background for professionalism
   • All service names must match official branding
3. Draft then Refine pattern:
   • Flash ($0.039) for initial draft
   • Review and adjust prompt
   • Pro ($0.134) for final version only after draft approved

Recommended Visual Set

Image	Audience	Focus
Master Architecture	Technical + Executive	Full OCI stack, all tiers
User Journey	Executive	4-5 step flow, business outcomes
Data Flow	Technical	Ingestion to Processing to Storage to Retrieval
Tier Comparison	Commercial	Side-by-side Basic/Advanced/Premium

Quality Gate 3: Visuals Compliant

• ZERO Oracle logos in any image
• Service names match official branding (26ai not 23ai)
• No spelling errors in diagrams
• Readable at presentation size (1920x1080 minimum)
• Architecture matches the SOLUTION-DESIGN.md

---

Phase 4: PROTOTYPE

Objective

Create an interactive HTML prototype the customer can experience.

Agent Strategy

For complex prototypes:

Task(subagent_type="coder")
"Build an interactive HTML prototype for [solution name]:
 Requirements:
 - Single index.html with embedded CSS/JS (no build tools)
 - Simulated [core workflow] with mock data
 - Processing states (loading animations, progress bars)
 - AI insights panel with simulated responses
 - Mobile responsive
 - PROTOTYPE banner clearly visible
 - Oracle Red (#C74634) accents, dark theme
 - NO real customer data -- all mock/synthetic"

For simple prototypes: Build directly in main context.

Prototype Quality Standards

• One core workflow, executed perfectly
• Every element has obvious purpose
• No placeholder content, no broken buttons
• Demo/prototype clearly labeled
• Mock data is realistic but obviously synthetic

Quality Gate 4: Prototype Functional

• Core interaction works end-to-end
• Processing states visible (loading, progress)
• No broken buttons or dead links
• PROTOTYPE/DEMO banner displayed
• No real customer data anywhere
• Mobile responsive (at minimum, not broken on mobile)

---

Phase 5: DELIVER

Objective

Package everything for customer presentation.

Output Structure

clients/[CODE]/
  SOLUTION-DESIGN.md           # Full SDD
  deliverables/
    images/
      architecture.png         # Master architecture (logo-free)
      user-journey.png         # Executive flow
      data-flow.png            # Technical detail
      tier-comparison.png      # Commercial comparison
    prototype/
      index.html               # Interactive demo
    docs/
      DISCOVERY.md             # Requirements captured
      BOM.md                   # Bill of Materials
  README.md                    # Project overview (codename only)

Final Confidentiality Audit

Before any delivery:

• grep for real customer names in all output files -- MUST return zero
• Solution name is generic or codename-based
• No internal Oracle pricing beyond published list prices
• No competitor attacks -- only OCI capability focus
• README.md contains ONLY: status, role, codename -- no industry/scope

Quality Gate 5: Delivery Ready

• All files in correct structure
• README complete with codename only
• Images are logo-free (visual inspection)
• Prototype loads and works standalone
• BOM pricing verified against official sources
• Would McKinsey present this to a Fortune 500 CEO? Yes/No

---

Quick Reference: Agent Routing Matrix

Need	Agent	Why Not Other?
OCI service research	Task(researcher)	Web search + fetch access
Competitor/market research	Task(researcher)	Web search access
Architecture design	Task(system-architect)	Architecture patterns expertise
Code review of prototype	Task(reviewer)	Code quality focus
Build prototype	Task(coder)	Implementation focus
Deep codebase exploration	Task(Explore)	Fast file search
Multi-step planning	Task(Plan)	Architecture planning
Pricing verification	Task(researcher)	Needs web access

Anti-Patterns

Do Not Do This	Do This Instead
Run all phases sequentially with no parallelism	Parallel research agents in Phase 1
Skip pricing verification	ALWAYS verify at oracle.com/cloud/price-list/
Include Oracle logos in visuals	Text labels only, Oracle Red accents
Persist client context to files	Conversation-only context, codename in files
Use claude-flow daemon for orchestration	Native Task agents -- zero infrastructure
Generate final visuals on first attempt	Flash draft first, Pro only after approval
Build complex prototype without coder agent	Delegate to Task(coder) for complex HTML/JS
Deliver without confidentiality audit	Always grep for real names before delivery

---

Integration with Other Skills

Phase	Complementary Skill	When to Invoke
DISCOVER	/oracle-research	Deep OCI service investigation
ARCHITECT	/oci-services-expert	Service selection + pricing
ARCHITECT	/oracle-sdd-generator	Full SDD document generation
VISUALIZE	/oracle-infogenius	Tiered image generation
VISUALIZE	/oracle-ai-architect-infogenius	Full validation pipeline
PROTOTYPE	/oracle-adk	If agent-based prototype needed
DELIVER	/oracle-confidentiality	Final audit before handoff

---

Version: 1.0 | Created: 2026-02-09 | Replaces: claude-flow swarm orchestration