crewai-agent-security

What problem does it solve?

CrewAI agent systems are vulnerable to role boundary violations, task delegation abuse, and shared-context poisoning, which can enable prompt injection and tool misuse.

Core Features & Use Cases

• Per-agent tool scoping enforces tools are bound to each agent's role to prevent cross-agent escalation.
• Context isolation and sanitization prevents leakage and instruction injection by stripping sensitive data before sharing outputs.
• Audit and logging of delegation, tool calls, and shared-context changes for post-incident analysis.
• Use cases include evaluating a CrewAI workflow for role adherence, validating tool permissions, and hardening agent collaborations in complex tasks.

Quick Start

Audit your CrewAI setup by listing each agent and confirming its allowed tools.