AgentShield AgentShield

AgentShield Security Studio

Inspect prompts and agent outputs locally, simulate tool firewall rules, and export Python SDK policy code.

0% cloud · zero telemetry
Load attack or benign preset
DecisionDENY
Risk LevelCRITICAL
Findings0
SanitizedYES

          

        


        

          

            

              
              
            

            

              
              
            

          

          

            
            
          

          

        


        

          

            ShieldedAgentRuntime Python snippet
            
          

          
from agentshield import ShieldedAgentRuntime, guard_tool

runtime = ShieldedAgentRuntime()
verdict = runtime.check_prompt("Summarize README.md", source="direct_user")

@guard_tool(runtime, tool_name="read_file")
def read_file(path: str) -> str:
    return open(path, encoding="utf-8").read()

        

      

    


    
  


  ",
    "benign": "Explain how Python asyncio task groups cancel sibling coroutines cleanly when a child raises ValueError."
  }