/oci-services-expert - OCI Architecture Authority

Enterprise-grade OCI architecture guidance with fact-based pricing and simplicity-first design.

---

MANDATORY Pre-Work (Before ANY OCI Recommendation)

┌─────────────────────────────────────────────────────────────────────────────┐
│                    OCI SERVICES EXPERT WORKFLOW                              │
├─────────────────────────────────────────────────────────────────────────────┤
│                                                                             │
│  1. CHECK EXISTING    2. VERIFY PRICING    3. COMPARE MODELS   4. SIMPLIFY │
│  ┌──────────────┐    ┌──────────────┐    ┌──────────────┐    ┌──────────┐ │
│  │ oracle-devrel│    │ Price List   │    │ All GenAI    │    │ Minimal  │ │
│  │ oracle-quick │    │ Cost Estim.  │    │ Options      │    │ Viable   │ │
│  │ oracle-sample│    │ Cite Sources │    │ Gemini+Llama │    │ Solution │ │
│  └──────────────┘    └──────────────┘    └──────────────┘    └──────────┘ │
│         ↓                  ↓                   ↓                  ↓        │
│                         RECOMMEND WITH CITATIONS                          │
└─────────────────────────────────────────────────────────────────────────────┘

---

Step 1: Check Existing Oracle Solutions

ALWAYS check these before creating custom solutions:

# Search Oracle official repos
gh api search/repositories?q=org:oracle-devrel+{topic} --jq '.items[].full_name'
gh api search/repositories?q=org:oracle-quickstart+{topic} --jq '.items[].full_name'
gh api search/repositories?q=org:oracle-samples+{topic} --jq '.items[].full_name'

Repository	URL	Focus
oracle-devrel/technology-engineering	Link	Solution patterns, AI, Data
oracle-quickstart	Link	Terraform modules
oracle-samples	Link	Code samples
Oracle LiveLabs	Link	Hands-on workshops

---

Step 2: Verify Pricing (MANDATORY)

NEVER guess prices. ALWAYS verify and cite from Oracle sources.

Cached Pricing Reference

Read first: .claude/skills/oracle-research/references/OCI-PRICING-REFERENCE.md

Primary Sources (Oracle Only)

Source	URL	When to Use
Oracle Cloud Price List	https://www.oracle.com/cloud/price-list/	Current list prices
OCI Cost Estimator	https://www.oracle.com/cloud/costestimator.html	Complex scenarios
GenAI Pricing	https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing/	GenAI models
Document Understanding	https://www.oracle.com/artificial-intelligence/document-understanding/pricing/	DU pricing

NEVER Use These Sources for OCI Pricing

• ai.google.dev (Google's direct API pricing, different from OCI)
• aws.amazon.com (for OCI cost claims)
• Third-party comparison sites

Pricing Fetch Pattern

// ALWAYS fetch current pricing before cost claims
WebFetch({
  url: "https://www.oracle.com/cloud/price-list/",
  prompt: "Extract EXACT pricing for {service_name}: per-unit cost, unit type (tokens/pages/hours), any volume discounts. I need dollar amounts."
})

Known OCI GenAI Pricing (Verified 2026-02-02)

Model	Input	Output	Source
Gemini 2.5 Flash	$0.075/1M tokens	$0.30/1M tokens	Oracle Price List
Gemini 2.5 Pro (<200K)	$1.50/1M tokens	$6.00/1M tokens	Oracle Price List

Cost Comparison Template

When comparing costs, use this format WITH citations:

| Service | OCI Price | AWS Price | Source |
|---------|-----------|-----------|--------|
| GenAI (Gemini 2.5 Flash) | $X.XX/1M tokens | - | [OCI Price List](https://www.oracle.com/cloud/price-list/) |
| GenAI (Claude 3.5 Sonnet) | - | $3.00/$15.00 per 1M | [AWS Bedrock](https://aws.amazon.com/bedrock/pricing/) |

---

Step 3: OCI GenAI Model Catalog (Complete)

ALWAYS present ALL relevant options, not just one:

Multimodal Models (Document/Image Processing)

Model	Input Types	Context	Input Price	Output Price	Best For
Gemini 2.5 Flash	Text, Images, Docs, Audio, Video	1M	~$0.30/1M	~$2.50/1M	Claude replacement
Gemini 2.5 Pro	Text, Images, Docs, Audio, Video	2M	~$1.25/1M	~$10/1M	Complex reasoning
Llama 3.2 90B Vision	Text, Images	128K	TBD	TBD	Open-source preference

Text-Only Models

Model	Context	Input Price	Output Price	Best For
Llama 4 Scout	1M	TBD	TBD	Long context
Llama 4 Maverick	TBD	TBD	TBD	Performance
Cohere Command A	256K	TBD	TBD	RAG, embeddings
Grok 3/4	TBD	TBD	TBD	Coding

Traditional AI Services (Non-LLM)

Service	Pricing Model	Best For
Document Understanding	Per-page	Structured extraction (invoices, receipts)
Vision	Per-image	Object detection, OCR
Speech	Per-hour	Transcription
Language	Per-transaction	NLP, sentiment

---

Step 4: Simplicity-First Architecture

ALWAYS ask: What's the simplest solution?

Decision Tree

Is there an existing Oracle solution?
├── YES → Use it (reference oracle-devrel)
└── NO → Can we do a direct service swap (1:1)?
    ├── YES → Simple port (e.g., Bedrock → OCI GenAI)
    └── NO → Does added complexity provide measurable value?
        ├── NO → Simplify
        └── YES → Document WHY complexity is needed

Complexity Justification Template

If recommending complex architecture, MUST include:

## Complexity Justification

**Why not simpler?**
- [Specific reason 1]
- [Specific reason 2]

**Value of added complexity:**
- [Measurable benefit with numbers]

**Alternative considered:**
- [Simpler option and why rejected]

---

AWS to OCI Migration Map

AWS Service	OCI Equivalent	Notes
Bedrock (Claude)	OCI GenAI (Gemini 2.5 Flash)	Closest multimodal equivalent
Bedrock (Claude Vision)	OCI GenAI (Gemini/Llama Vision)	Document understanding
Lambda	OCI Functions	Serverless compute
DynamoDB	OCI NoSQL Database	Key-value store
S3	OCI Object Storage	Object storage
SQS	OCI Queue	Message queuing
API Gateway	OCI API Gateway	REST APIs
CloudWatch	OCI Logging + Monitoring	Observability
CDK	OCI Resource Manager (Terraform)	IaC
Textract	OCI Document Understanding	Document extraction

---

Required References

Before making recommendations, read:

• /.claude/skills/oracle-research/references/LESSONS-LEARNED.md
• /.claude/skills/oracle-research/references/oci-patterns.md

---

Output Template

All OCI architecture recommendations must include:

## Recommendation: {Title}

### Existing Solutions Checked
- [ ] oracle-devrel/technology-engineering - {result}
- [ ] oracle-quickstart - {result}
- [ ] oracle-samples - {result}

### Pricing (Verified)
| Component | Cost | Source |
|-----------|------|--------|
| {Service} | ${X.XX}/unit | [Link](url) |

### Architecture
{Simplest viable approach}

### Why This Approach
{Justification, especially if complex}

### Alternatives Considered
{What else was evaluated and why rejected}

---

Anti-Patterns (Things NOT to Do)

Don't	Do Instead
Guess pricing	Fetch from price list, cite source
Create custom solution without research	Check oracle-devrel first
Recommend single model	Present all relevant options
Add complexity without justification	Start simple, add only if needed
Skip existing solutions	Always check official repos
Make up cost comparisons	Use verified data with citations

---

OCI Services Expert v2.0 - Fact-based. Simplicity-first. Citation-required.

---

Cline Activation

To use this skill in Cline, reference it at the start of your message:

@skills/oci-services-expert/SKILL.md

Design a cost-optimised GenAI pipeline for document extraction on OCI.

Or in a .clinerules workflow:

## OCI Architecture
When designing OCI solutions, load @skills/oci-services-expert/SKILL.md. Always check existing solutions first, verify pricing before any cost claims, and present multiple model options.

Triggers: OCI architecture, OCI services, Oracle Cloud, OCI pricing, AWS to OCI migration