examples

ZIRAN Examples

Hands-on examples that show how to test different AI agent architectures with ZIRAN.

Quick start

cd examples/

# install the examples workspace (uses uv)
uv sync                        # base examples (no API key needed)
uv sync --extra langchain      # + LangChain examples
uv sync --extra crewai         # + CrewAI example
uv sync --extra rag            # + RAG / FAISS examples
uv sync --extra all            # everything

# run any example
cd 01-static-analysis
./run.sh

Tip: Copy .env.example to .env and fill in your API keys before running examples 09–14.

---

No API key required

These examples use ZIRAN's built-in scanner without calling any LLM.

#	Example	What it demonstrates
01	Static Analysis	Analyse raw Python source code for security issues
02	Attack Library	Load the built-in attack library and generate custom vectors from YAML
03	Dynamic Vectors	Create dynamic attack vectors with inline YAML
04	Skill CVE	Query the embedded CVE knowledge base
05	PoC Generation	Generate a proof-of-concept exploit from a CVE entry
06	Policy Engine	Define a YAML security policy and evaluate findings against it
07	CI/CD Quality Gate	Fail a build when findings exceed a YAML-configured threshold
08	Custom Adapter	Implement AgentAdapter for any agent framework

LLM scans (API key required)

These examples call an LLM provider. Set OPENAI_API_KEY in ../.env first.

#	Example	Agent architecture	Extra deps
09	LangChain Scan	ReAct agent with calculator + search tools	--extra langchain
10	Vulnerable Agent	Intentionally weak HR chatbot (finds vulns!)	--extra langchain
11	RAG Financial Advisor	FAISS-backed advisor with confidential client data	--extra rag
12	Router RAG	Dynamic router → knowledge base / customer DB / market API	--extra rag
13	Supervisor Multi-Agent	Supervisor delegates to HR, Finance, IT sub-agents	--extra langchain
14	CrewAI Scan	CrewAI research crew (native adapter, no LangChain)	--extra crewai
15	Remote Agent Scan	Scan agents over HTTP (REST, OpenAI, MCP, A2A)	--extra remote
16	LLM-as-a-Judge	Enhanced detection with AI-powered judge (multi-provider)	--extra langchain + litellm
17	Bedrock Agent Scan	Scan an Amazon Bedrock Agent via the AWS SDK	--extra bedrock
18	AgentCore Scan	Scan an AgentCore-deployed agent in-process (mock included)	--extra agentcore

---

Folder structure

Every example lives in its own folder with:

NN-example-name/
├── main.py          # the example script
├── run.sh           # one-click runner (checks env, launches uv run)
├── README.md        # what it does, architecture, expected results
└── *.yaml / *.py    # config or sample files (where needed)

Reports are written to NN-example-name/reports/ (git-ignored).