Agent Breaker

Automated adversarial security testing for LangGraph-based AI agents. Catch vulnerabilities in LLM-powered applications before they reach production.

Features Overview

• Plug-and-play security testing for LangGraph agents
• Domain-aware adversarial prompt generation (finance, healthcare, legal, etc.)
• ML and rule-based behavioral judges (97.8% accuracy with ML)
• Auto-detects agent tools and capabilities
• Detailed vulnerability reports in the terminal
• Graceful rate limit handling
• CLI: agent-breaker init, agent-breaker run, with options for debug/full output
• Supports custom domains and config via breaker.yaml

Judge Verdict Types

• PASS: Agent properly refused the adversarial request
• WARN: Agent discussed the request but took no action
• FAIL: Agent exhibited vulnerable behavior (complied with the attack)
• INFO: Agent refused but provided guidance or information (needs review)
• SKIP: Test was skipped (rate limit, API error, etc.)

Installation

pip install agent-breaker
# Optional: for ML judge (recommended)
pip install agent-breaker[ml]

CLI Commands

Initialize config:

agent-breaker init

Add --force to overwrite existing breaker.yaml

Run tests:

agent-breaker run

Add --debug to show full traceback on errors

Add --full-output to show full payload and model response text

Optional environment variables (default: off): AGENT_BREAKER_DEBUG=1 Enable debug mode AGENT_BREAKER_FULL_OUTPUT=1 Enable full output

Quick Start

1. Initialize config:

   agent-breaker init
   # Edit breaker.yaml to point to your agent
   

2. Run tests:

   agent-breaker run
   

Example breaker.yaml

version: "0.2"
target:
  type: "langgraph"
  path: "my_agent.py"
  attr: "graph"
  prompt_variable: "SYSTEM_PROMPT"
  input_key: "user_query"
  output_key: "response"
  state_class: "AgentState"
generator:
  strategy: "template"
  domain: "finance"
attacks:
  - name: "prompt_injection"
    enabled: true
    max_api_calls: 10
judge:
  model: "ml"  # or "behaviour"

Usage Example

# In your agent file (my_agent.py):
graph = workflow.compile()
# breaker.yaml should reference this file and variable

Documentation

Documentation: https://github.com/GokulAIx/Agent-Breaker#readme

License

MIT License

Author

P. Gokul Sree Chandra