Diagram generation: ref-arch-driven procedure + spec validator + KB enrichment

The diagram path now follows a documented standard procedure (lookup
the closest Oracle Architecture Center reference → confirm components
→ author absolute_layout → spec validator → render → visually verify)
and ships persistent guardrails so layout regressions can't recur.

Persistent procedure changes (apply to all users, all sessions):
- tools/diagram_spec_validator.py — geometry checks (CONTAINER_TOO_THIN,
  CONTAINER_PADDING_VIOLATION, LABEL_OVERFLOW_PARENT) run BEFORE either
  renderer (drawio + PPTX). Catches the subnet-collapse / label-overflow
  bugs that the post-render drawio validator missed.
- tools/oci_diagram_gen.py + tools/oci_pptx_diagram_gen.py — call the
  spec validator before emitting any output. Adds mysql / mysql_heatwave
  type aliases.
- tools/archcenter_pattern_lookup.py — scores against cached page
  descriptions (not just the 1-line summary), supports --queries for
  multi-fragment composition, and applies synonym expansion via
  kb/architecture-center/synonyms.yaml so "LB HA cross AD" matches
  "load balancer high availability availability domain".
- kb/architecture-center/synonyms.yaml — canonical synonym table
  (load balancer, autonomous database, data guard, …) used by the
  lookup scorer.

KB enrichment:
- tools/archcenter_description_fetcher.py + 121 cached _description.md
  under kb/diagram/assets/archcenter-refs/<slug>/. Removes the runtime
  dependency on docs.oracle.com when authoring specs and feeds the
  pattern-lookup scorer.
- 110+ cached .drawio / .svg / .png references for offline reuse,
  plus the OCI Toolkit v24.2 import (kb/diagram/assets/oci-toolkit-drawio).

Documentation:
- docs/skill/output-formats.md — new "Standard diagram-generation
  procedure (MANDATORY)" + geometry rules + the new validator entry.
- SKILL.md option 2 — references the mandatory procedure.
- README.md — describes the spec validator, archcenter_pattern_lookup
  and description fetcher, and updates the KB-health table.

Tooling that backs the procedure (cumulative across recent sessions):
tools/archcenter_case_runner.py, archcenter_batch_driver.py,
archcenter_zip_downloader.py, drawio_visual_validator.py,
drawio_fidelity_eval.py, harvest_drawio_icon.py, import_oci_library.py,
oci_pptx_diagram_gen.py, oci_pptx_render.py, refresh_pptx_icon_index.py.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

kb/diagram/assets/archcenter-refs/ai-fraud-detection/_description.md

@@ -0,0 +1,158 @@
+# Deploy a multi-agent AI fraud detection system on OCI
+
+- Source: https://docs.oracle.com/en/solutions/ai-fraud-detection/index.html
+- Date: 2025-06
+- Type: reference-architecture
+- Services: genai, streaming, adb-s
+- Tags: ai-ml, autonomous
+
+## Summary (catalog)
+
+Multi-agent AI system for real-time fraud detection. Streaming ingestion of transaction data, ML models for anomaly detection, GenAI agents for investigation and decision support.
+
+## Architecture (fetched from source)
+
+Architecture
+
+ 
+
+ 
+This architecture shows a multi-agent AI fraud detection system on Oracle Cloud
+ Infrastructure (OCI).
+ 
+
+ 
+This design uses multiple AI agents to provide key insights, gather evidence,
+ and produce a comprehensive fraud analysis.
+
+ 
+At the core is a model context protocol (MCP) server that orchestrates agent
+ interactions. Specialized agents handle distinct tasks. For example, a data retrieval
+ agent queries enterprise data sources and a fraud analyzer agent evaluates and explains
+ anomalies. The flow begins when an event, such as a suspicious transaction alert or an
+ investigator’s query, triggers the orchestrator. The orchestrator then delegates
+ subtasks to the agents and consolidates their results by using a fan-in and fan-out
+ design pattern. Each agent uses OCI-native services to perform tasks (database queries,
+ LLM inference, and so on), and the orchestrator translates between agent and Oracle
+ system contexts, ensuring that each agent gets the information it needs in the format it
+ expects. 
+
+ 
+The following diagram shows process flow overview:
+
+ 
+ Description of the illustration ai-fraud-detection-flow.png 
+
+ ai-fraud-detection-flow-oracle.zip 
+
+ 
+ 
+- MCP orchestrator server 
+The model context protocol
+ (MCP) server is the coordinating hub that orchestrates agent actions and
+ maintains the overall context or state of an investigation. It uses MCP to
+ standardize how agents invoke tools and exchange data. Acting as a “central
+ brain,” it receives the initial request (fraud alert or analysis query) and
+ calls the appropriate agents in sequence. It also translates high-level agent
+ intents into low-level operations on Oracle systems, such as converting an
+ agent’s request for customer information into an SQL query, and converting SQL
+ results into a natural language response. This approach decouples agents from
+ direct system calls by using the orchestrator as a bridge between the agent
+ logic and enterprise data, enabling flexible updates and centralized control. In
+ the first phase of this architecture, the server is a lightweight server derived
+ from Google’s Gen AI Toolbox running on Oracle Cloud Infrastructure
+ Compute or OCI Kubernetes Engine .
+ 
+
+ 
+ 
+- Data retrieval agent 
+The data retrieval agent is
+ a specialized agent responsible for fetching relevant data from enterprise
+ sources. For example, upon receiving a customer ID or transaction ID from the
+ orchestrator, it queries the Oracle Autonomous Database or other OCI data stores for information such as recent transactions, account
+ profiles, claims history, and so on. You can implement this agent by using OCI Functions (serverless) to call tools hosted on an MCP server for Autonomous Database . The agent contains all data-access logic. The orchestrator server may use a
+ predefined tool for this agent, such as a YAML-configured
+ LookupTransaction or GetCustomerProfile 
+ tool that knows how to run the proper SQL on the Autonomous Database . Similar to how Google Gen AI Toolbox uses YAML-defined tools to let agents
+ perform database operations, this design defines database queries as
+ configuration-driven tools. In the first phase, the data agent simply executes
+ these queries without AI decision-making involvement and returns the results to
+ the orchestrator.
+ 
+
+ 
+ 
+- Fraud analyzer agent 
+The fraud analyzer agent is
+ the cornerstone agent that assesses the data for signs of fraud and generates
+ insights. This agent ingests the context, such as the transaction details,
+ customer info, or historical patterns provided by the orchestrator and applies
+ AI/ML logic to determine if the scenario is likely fraudulent. In phase 1, this
+ could be a rule-based engine or an OCI Anomaly Detection model to provide a
+ quick, deterministic response. For example, it might flag anomalies such as a
+ transaction far outside normal range or multiple claims in a short time span.
+ The agent then produces a fraud score or classification and possibly an
+ explanation. 
+
+ 
+In phase 2, the fraud analyzer agent is
+ augmented with LLM capabilities by using OCI Generative AI or Oracle pretrained models to generate human-readable investigative
+ narratives. In this way, generative AI automatically creates a concise report of
+ the findings, summarizing why a transaction was flagged and referencing the data
+ directly such as how a customer’s recent transactions show unusual high-value
+ purchases overseas, which deviates from their normal pattern by 5σ (5 sigma),
+ indicating high fraud likelihood. Oracle’s own Financial Services division has
+ highlighted the value of such generative narratives in accelerating
+ investigations. In phase 2, the fraud analyzer agent can use an OCI LLM to both
+ analyze the data and explain the results. For example, it might use a prompt
+ that incorporates the data and asks the model to analyze the fraud risk, or it
+ could perform tool-assisted reasoning by first calling a calculation tool, and
+ then having the LLM elaborate on the results.
+ 
+
+ 
+ 
+- Additional agents (as needed) 
+The architecture
+ supports the ability to plug in other agents to enrich the analysis. For
+ example, an external check agent could call third-party services, such as
+ sanctions lists or credit bureaus, to gather more evidence on the entity
+ involved. Another could be a notification and case management agent that, after
+ fraud is confirmed, logs the case in a system or triggers an alert to a human
+ investigator. The orchestrator’s ability to manage multiple agents and
+ coordinate complex workflows allows new agents to be added without disturbing
+ existing ones. This modularity makes the system extensible for demo showcases
+ that can start with two agents and later attach more for other demo scenarios
+ such as compliance checks, customer messaging, and so on.
+
+ 
+ 
+ 
+
+ 
+
+ 
+
+ 
+
+ 
+
+ 
+ Title and Copyright Information 
+
+ 
+
+ 
+Deploy a multi-agent AI fraud detection system on OCI
+
+ 
+G34124-01
+
+ 
+June 2025
+
+ 
+ Copyright © 2025, 
+
+Oracle and/or its affiliates.

kb/diagram/assets/archcenter-refs/ai-fraud-detection/readme.txt

@@ -0,0 +1,9 @@
+The diagram you downloaded is available in these formats:
+- DRAWIO
+- SVG
+
+You can customize them for your organization using the associated tools:
+- For DRAWIO format, use draw.io for Confluence, online at diagrams.net, or the desktop app. Go to diagrams.net for more information.
+- For SVG format, use an SVG editor such as Inkscape or Sketsa SVG Editor, which are free and available for Windows, macOS, Linux.
+
+Note that all diagram components are ungrouped and in a single layer.