Usage Examples

Effective Prompting

For reliable results, always state three things clearly at the start of your session:

1. Authorization — confirm you are authorized to test the target
2. Ownership / scope — state what you own or what the program scope covers
3. Objective — what you want to find or test

---

Example Prompts

Penetration Test — Full Scope

I am an authorized security researcher. My company owns example.com and all its
subdomains. I want a full authorized penetration test: start with passive recon,
then active scanning, web vulnerability discovery, and summarize findings with
recommended next steps.

Bug Bounty — Web Application

I am participating in the HackerOne bug bounty program for example.com. The scope
covers *.example.com. Please start with subdomain enumeration, httpx probing, then
run nuclei and dalfox against live targets. Flag anything that looks like a valid
finding within scope.

CTF Challenge

I am solving a CTF challenge. The target binary is at /tmp/challenge. Please run
checksec, then strings, then analyze with radare2 and suggest an exploitation approach.

WiFi Pentest (authorized lab)

I am testing my own home WiFi lab. The interface is wlan0. Please put it into monitor
mode, capture a WPA2 handshake from BSSID AA:BB:CC:DD:EE:FF, and attempt to crack it
using rockyou.txt.

Cloud Security Assessment

I have AWS credentials configured for my company's test account. Please run prowler
against the account to check for misconfigurations and compliance issues, then
summarize critical and high findings.

---

Switching Between Agent and Manual Mode

NyxStrike supports switching between AI-driven and manual execution within the same session:

1. Start a session and let the AI plan and execute the attack chain.
2. At any point, open the dashboard and switch to Manual Tool Execution.
3. Use the workbench to run individual tools, review outputs, and chain artifacts manually.
4. Return to the AI agent — it picks up from the current session state.

---

Compact Mode (small/local LLMs)

For lightweight or local LLMs that cannot handle a large tool context, use --compact:

nyxstrike-env/bin/python3 nyxstrike_mcp.py \
  --server http://127.0.0.1:8888 \
  --compact

In compact mode, the agent has access to two tools only:

• classify_task — describes what to do given a target and objective
• run_tool — executes any tool by name with parameters

The LLM uses these two tools to orchestrate the full arsenal without needing every tool registered individually in its context.