tyga-crewai 0.1.2

Safety shield for CrewAI — screen agent tool calls before execution

tyga-crewai

Safety shield for CrewAI. Drop-in SafeShellTool that screens commands before execution.

Install

pip install tyga-crewai

Quick Start

from crewai import Agent, Task, Crew
from tyga_crewai import SafeShellTool

agent = Agent(
    role="DevOps Engineer",
    goal="Monitor and maintain server health",
    tools=[SafeShellTool()],
)

task = Task(
    description="Check disk usage and running services.",
    expected_output="Summary of system health.",
    agent=agent,
)

crew = Crew(agents=[agent], tasks=[task])
crew.kickoff()

# Safe commands (df -h, systemctl status) run normally
# Dangerous commands (rm -rf, chmod 777) are blocked

How it works

1. Gate 1 (local): Regex denylist screens commands in <5ms. Free, no API key needed.
2. Gate 2 (cloud): Set A2A_API_KEY env var to enable LLM judge evaluation via a2ainfrastructure.com.

Get an API key

Sign up free at a2ainfrastructure.com. 500 evaluations/month on the free tier.

Links

• Quick Start Guide
• API Docs

License

Proprietary — Tyga.Cloud Ltd. See LICENSE file.