Module Reference

Every Python module currently in the repository, with its purpose (taken from module docstrings). The live, machine-generated version is Module Inventory; the import graph is Module Dependency Graph.

capstone_agent/ — ADK agent backend

Module	Purpose
`agent.py`	Root agent definition — entry point for `adk run capstone_agent`; imports all modules and wires the root orchestrator plus 21 specialist LLM agents
`app.py`	ADK `App` wrapper — plugins, compaction, resumability around the root agent
`a2a_server.py`	Expose the agent over the Agent2Agent (A2A) protocol (Day 5a)
`llm.py`	Model registry and factory — the one place model selection happens (Model Registry)
`orchestration.py`	Multi-agent orchestration building blocks — the 3 pipeline factories, agent-as-tool, code exec, remote A2A
`tools.py`	Clinical tool definitions (production) — Pydantic-validated, clinical event logging
`callbacks.py`	Security callbacks — defense in depth (Security Layers)
`security.py`	PII detection, secret scanning, input sanitization (pure functions)
`memory.py`	Memory management — Layers 2, 3, 4 (Memory Layers)
`context.py`	Context engineering — token budgeting, compaction, structured assembly (Layer 1)
`observability.py`	Structured logging, tracing, and audit trail (Observability)
`plugins.py`	Observability plugins (Day 4a) — ObservabilityPlugin, ClinicalAuditPlugin, LoggingPlugin
`prompts.py`	Agent instruction templates (under 60 lines each)
`models.py`	Pydantic models for tool input/output contracts (ToolResponse, ToolError)
`config.py`	Centralized configuration and secret management (redaction, clinical governance)
`clinical_schemas.py`	Clinical database schema definitions and query engine (SQL DDL + validation)
`database.py`	SQLite database layer — real persistence for all clinical data
`document_processor.py`	Document processing — real PDF and image extraction using PyMuPDF and Gemini
`mock_data.py`	Deterministic mock data for Clinical AI Kit
`human_in_the_loop.py`	Human-in-the-loop / long-running operations (Day 2b)

clinical_app/ — FastAPI product server

Module	Purpose
`app.py`	FastAPI routes for the clinician-facing application; serves the frontend build
`agent_runtime.py`	Deterministic product adapters for the real clinical agent tool layer
`live_bridge.py`	Lazy Google ADK execution bridge for live product mode
`repository.py`	Session-isolated mutable repository for deterministic product demos
`models.py`	Pydantic contracts for the clinician product API
`document.py`	Document parsing and upload policy for clinical evidence files
`system.py`	Real system health checks, latency monitoring, and database seeding for the real tenant

mcp_server/

Module	Purpose
`server.py`	Clinical MCP server — real database-backed tools via Model Context Protocol (FastMCP)

scripts/ — harness and showcase utilities

Module	Purpose
`check_harness.py`	PreToolUse hook — verifies harness integrity (Claude Harness)
`sync_harness.py`	Mirrors `.claude/` → `.agents/` and CLAUDE.md → AGENTS.md
`sync_wiki.py`	Stop hook — deterministic wiki/harness auto-update (Development Workflow)
`export_diagrams.py`	Export Project Wiki draw.io diagrams into frontend public assets
`generate_database_showcase.py`	Generate a large governed SQLite cohort for the database agent
`generate_extraction_showcase.py`	Generate synthetic extraction assets for the image extraction agent
`generate_multimodal_patient_showcase.py`	Generate multimodal patient bundles for the Q&A agent

Other top-level areas

Area	Purpose
`frontend/`	React/Vite/TypeScript clinical UI — 16 core screens plus 3 documentation/utility routes; `dist/` served by FastAPI
`eval/`	ADK evaluation suite — `capstone.evalset.json` + `test_config.json` (Testing and Eval)
`tests/`	pytest + async test suite (Testing and Eval)
`deployment/`	Dockerfile, cloudbuild.yaml, Agent Engine config (Deployment)
`docs/`	Architecture and product documentation