Claude Agents Integration

Overview

ClaudeAgentGuard wraps Claude Agent SDK client loops to validate outputs before they are returned to users or trigger tool calls.

Tier: Team+

Why It Matters

Claude Agents run autonomous tool-use loops. Governance must intercept loop outputs without breaking the agent's tool execution flow.

Prerequisites

pip install agentrust-sdk
# Team+ API key
# Claude Agents SDK installed in your environment

Step-by-Step Guide

from agentrust_sdk import ClaudeAgentGuard

guard = ClaudeAgentGuard(agent_id="claude-agent")

# Wrap client or agent loop
with guard.wrap(client):
    result = client.run(prompt="Analyze this document")

The guard validates each loop iteration's output through the AgentTrust gateway.

Examples

See agentrust_sdk/agentrust_sdk/adapters/claude_agents.py for implementation details.

Best Practices

• Wrap at the client level for consistent coverage
• Use block_on_block=True for sensitive document processing
• Combine with PII policy pack for customer data

Common Mistakes

• Team tier required — OSS key will not activate adapter
• Wrapping too late (after output already delivered)

Troubleshooting

Related Resources

• OpenAI Agents
• Tiers and Capabilities