feat: add CrewAI + SidClaw governance integration

[VladUZH]

Summary

Adds a self-contained integration example showing how to wrap CrewAI agents with SidClaw — an open-source approval and audit layer for AI agents (Apache 2.0 SDK, PyPI: sidclaw).

The example is a DevOps crew with three governed tools, each demonstrating a different policy outcome:

Tool	Classification	SidClaw decision
check_service_health	internal	Allow — runs immediately, outcome traced
deploy_to_production	confidential	Approval required — waits for human sign-off in dashboard
run_db_migration	restricted	Deny — blocked by policy, tool never executes

What's included

• integrations/CrewAI-SidClaw/
  • README.md — setup guide, expected output, how it works
  • crew.py — governed DevOps crew with three tools and different risk profiles
  • main.py — runnable entry point
  • requirements.txt — crewai, sidclaw, python-dotenv
  • .env.example — environment variable template

Integration pattern

from sidclaw import SidClaw
from sidclaw.middleware.crewai import govern_crewai_tool

client = SidClaw(api_key=os.environ["SIDCLAW_API_KEY"], agent_id="devops-crew")

governed_deploy = govern_crewai_tool(
    DeployToProductionTool(),
    client=client,
    data_classification="confidential",
)

govern_crewai_tool() wraps BaseTool._run. Before execution: SidClaw policy engine evaluates the action (< 50ms). After: outcome is recorded to a hash-chain audit trace.

Why this is useful for CrewAI users

CrewAI agents that call deploy_to_production or run_db_migration take irreversible actions. This integration adds:

• Human approval checkpoints before high-risk tool calls
• Policy-based blocking (deny rules prevent restricted actions from running at all)
• Tamper-proof audit trail for every tool execution

The SDK is Apache 2.0. No credential changes required — SidClaw wraps existing tools in one line.

Testing

pip install -r requirements.txt
cp .env.example .env  # add your keys
python main.py

Requires a SidClaw API key (free tier at app.sidclaw.com covers 5 agents).