AI Governance Assistant

An interactive, multi-phase conversational agent that guides developers, architects, and program managers through a structured AI governance risk assessment. Built with the Anthropic SDK and callable directly from the Claude Code terminal.

What it does

The assistant conducts a three-phase governance review of any AI service or feature:

Phase	Name	What happens
1	Data Collection & Context Mapping	The agent interviews you about your AI service — what it does, what data it uses, who it affects, and how autonomous it is
2	Risk Analysis & Vulnerability Assessment	The agent analyzes your answers across bias & fairness, privacy & data protection, security & robustness, and regulatory alignment
3	Mitigation & Governance Recommendations	A prioritized, structured report with technical mitigations, process governance steps, transparency guidance, and an incident response protocol

After the report, you are asked whether you are satisfied. Answering no restarts the full assessment from Phase 1.

Prerequisites

• Python 3.9+
• An Anthropic API key

Setup

# 1. Navigate to the project directory
cd AiGovernanceAssistant

# 2. Install dependencies
pip install -r requirements.txt

# 3. Set your Anthropic API key
export ANTHROPIC_API_KEY="sk-ant-..."

Running the agent

Option A — directly from the terminal

python3 governance_assistant.py

Option B — from the Claude Code terminal (slash command)

Open Claude Code in this project directory and type:

/project:governance-assistant

Claude will display the setup instructions and the command to launch the agent.

Option C — as a Claude Code sub-agent

Claude Code will automatically use .claude/agents/governance-assistant.md when you ask it to perform a governance review within a Claude Code session via the Agent tool.

Example session flow

The agent greets you and begins Phase 1 immediately:

Assistant: Hello! I'm your AI Governance Assistant...
           Let's start with Phase 1. Can you describe the AI service
           or feature you're building and the problem it solves?

You: We're building a loan approval model for retail banking...

[Phase 1 Q&A continues until all 6 areas are covered]

Assistant: Thank you — I now have everything I need. Moving to Phase 2...

[Phase 2 runs autonomously — no input required]

Assistant: Phase 2 complete. Generating your governance report...

[Phase 3 structured report is printed]

────────────────────────────────────────────────
Are you satisfied with this report? (yes / no): yes

Thank you for using the AI Governance Assistant. Goodbye!

Project structure

AiGovernanceAssistant/
├── governance_assistant.py       # Main agent script
├── requirements.txt              # Python dependencies
├── CLAUDE.md                     # Agent persona & workflow (read by Claude Code)
└── .claude/
    ├── agents/
    │   └── governance-assistant.md   # Sub-agent definition for Claude Code
    └── commands/
        └── governance-assistant.md   # /project:governance-assistant slash command