# Topic 5: Inter-File Connections (Call Graph)

**Audit Date**: 2026-02-01
**Auditor**: Agent Antigravity
**Scope**: Call Dependency Analysis

---

## 1. Primary Execution Chain (The "Happy Path")

### A. Ingress (API Layer)
*   **`app/main.py`**
    *   Imports `app/api/routes.py` -> `api_router`
    *   Imports `app/agents/orchestrator.py` -> `orchestrator.initialize()`
*   **`app/api/routes.py`**
    *   Calls `app/utils/guvi_handler.py` -> `GUVIHandler.process_guvi_message()`
    *   Dependencies: `settings.GUVI_API_KEY`, `GUVIInputRequest` schema.

### B. Logic Controller (The Orchestrator)
*   **`app/utils/guvi_handler.py`**
    *   Calls **`app/agents/orchestrator.py`** -> `Orchestrator.process_message()`
    *   Calls **`app/utils/callback_client.py`** -> `guvi_callback.send_final_result()`
*   **`app/agents/orchestrator.py`**
    *   Calls **`scam_detector.py`** -> `detect()`
    *   Calls **`intelligence_extractor.py`** -> `extract()`
    *   Calls **`conversation_manager.py`** -> `get_or_create()`, `update()`
    *   Calls **`adaptive_strategy.py`** -> `analyze_scammer_behavior()`
    *   Calls **`persona_engine.py`** -> `generate_response()`
    *   Calls **`threat_engine.py`** -> `analyze()`

### C. The Generators (Core Logic)
*   **`app/agents/scam_detector.py`**
    *   Calls **`app/core/llm_client.py`** -> `generate_with_cot()`
    *   Calls **`app/intelligence/emotional_analyzer.py`** -> `analyze()`
*   **`app/agents/persona_engine.py`**
    *   Calls **`app/core/llm_client.py`** -> `generate()`
    *   Calls **`app/decoys/victim_profiles.py`** -> `profile_generator.generate_profile()`
    *   Calls **`app/core/engagement_delay.py`** -> `simulate_typing()` (Simulated Latency)

---

## 2. Intelligence & Persistence Links

### A. Intelligence Chain
*   **`app/agents/orchestrator.py`**
    *   Calls **`app/intelligence/scammer_profiler.py`** -> `analyze_behavior()`
    *   Calls **`app/intelligence/graph_threat_intel.py`** -> `add_intelligence()`
    *   Calls **`app/intelligence/risk_scorer.py`** -> `calculate_risk_score()`

### B. Persistence Chain
*   **`app/agents/conversation_manager.py`**
    *   Calls **`app/database/db.py`** -> `get_session()`, `commit()`
    *   Calls **`app/database/models.py`** -> `Conversation`, `Message` tables.

---

## 3. Dependency Visualization

```mermaid
graph TD
    Main[main.py] --> Routes[api/routes.py]
    Routes --> Handler[utils/guvi_handler.py]
    Handler --> Orchestrator[agents/orchestrator.py]
    
    subgraph Agents
        Orchestrator --> Detector[scam_detector.py]
        Orchestrator --> Extractor[intel_extractor.py]
        Orchestrator --> Strategy[adaptive_strategy.py]
        Orchestrator --> Persona[persona_engine.py]
    end
    
    subgraph Intelligence
        Orchestrator --> Threat[threat_engine.py]
        Orchestrator --> Scorer[risk_scorer.py]
        Detector --> Emotional[emotional_analyzer.py]
    end
    
    subgraph Core
        Detector --> LLM[core/llm_client.py]
        Persona --> LLM
        Extractor --> LLM
    end
    
    Handler -.-> Callback[utils/callback_client.py]
```

## 4. Key Findings
*   **Centralization**: `Orchestrator` is the absolute center of the graph. No sub-agent calls another sub-agent directly; they all return to the Orchestrator. This prevents "Spaghetti Code".
*   **Isolation**: `api/routes.py` never imports `llm_client` directly. It MUST go through `guvi_handler` -> `orchestrator`.
*   **Utility Independence**: `callback_client.py` is independent of the agents, ensuring reporting works even if agents crash.