phantom-agent 0.8.4

Autonomous Offensive Security Intelligence — AI-powered penetration testing

☠ PHANTOM

Autonomous Offensive Security Intelligence

" Why So Serious ?! "

AI-powered multi-agent penetration testing that thinks like a hacker.

Quick Start · Features · Architecture · Documentation · Contributing

---

📖 Table of Contents

• Overview
• Features
• Architecture
• Quick Start
• Usage
• Configuration
• CI/CD Integration
• Development
• Documentation
• Contributing
• License

🎯 Overview

Phantom deploys autonomous AI agents that act like expert penetration testers — they probe your systems dynamically, discover vulnerabilities, chain attack paths, and validate findings through real proof-of-concepts.

Unlike static scanners that just pattern-match, Phantom thinks: it reads responses, adapts strategy, exploits chained vulnerabilities, and proves every finding with working PoCs. No false positives. No manual triage.

Built for security teams, bug bounty hunters, and developers who need fast, accurate offensive testing without the overhead of manual pentesting.

Why Phantom?

Traditional Scanners	Phantom
Pattern matching / signatures	AI reasoning + adaptive strategy
High false positive rate	Every finding validated with PoC
Single-pass scanning	Multi-agent collaboration & chaining
Generic reports	MITRE ATT&CK mapped + compliance ready
Manual triage needed	Actionable findings, remediation steps

✨ Features

• 🤖 Fully Autonomous — AI agents that think, adapt, and act like real pentesters
• 🔬 20+ Security Tools — nmap, nuclei, sqlmap, ffuf, semgrep, katana, and more in a sandboxed Docker environment
• 🛡️ Secure Sandbox — All operations inside isolated Docker containers. Zero risk to your host
• 👥 Multi-Agent System — Specialized agent trees: discovery → exploitation → validation → reporting
• 🎯 Real PoCs — Every vulnerability comes with working proof-of-concept code
• 📊 MITRE ATT&CK Mapping — Findings automatically mapped to TTPs and techniques
• 📋 Compliance Reports — OWASP Top 10, PCI DSS, SOC 2 mapping out of the box
• 📦 SARIF Output — Native GitHub Security tab integration
• 🔄 Differential Scanning — Track new/fixed vulnerabilities across scan runs
• 🧠 Knowledge Persistence — Learns from past scans, avoids redundant work
• 🔔 Notifications — Webhook & Slack alerts on critical findings
• 🔌 Plugin System — Extend with custom tools and workflows
• ⚡ Multiple LLM Providers — OpenAI, Anthropic, Google, Groq, Ollama, and more via LiteLLM

🏗️ Architecture

System Overview

flowchart TB
    classDef person fill:#08427B,stroke:#073B6F,color:#fff
    classDef system fill:#9b59b6,stroke:#8e44ad,color:#fff
    classDef external fill:#555555,stroke:#333333,color:#fff

    user["👤 Security Engineer"]

    phantom["☠ PHANTOM<br/>Autonomous Pentest Agent"]

    target["🎯 Target System<br/>(Web App / API / Network)"]
    llm["🧠 LLM Provider<br/>(OpenAI / Anthropic / Groq / Ollama)"]
    sandbox["🐳 Docker Sandbox<br/>(Kali Tools Environment)"]

    user --> |"Configure & Launch"| phantom
    phantom --> |"Tests"| target
    phantom --> |"Reasons & Plans"| llm
    phantom --> |"Executes Tools"| sandbox
    phantom --> |"Reports"| user

    class user person
    class phantom system
    class target,llm,sandbox external

Agent Architecture

flowchart TD
    subgraph Orchestrator["☠ Root Agent (Orchestrator)"]
        direction TB
        A[Analyze Target] --> B[Create Strategy]
        B --> C[Spawn Specialist Agents]
    end

    subgraph Discovery["🔍 Discovery Agents"]
        D1[Port Scanner]
        D2[Web Crawler]
        D3[Tech Fingerprint]
    end

    subgraph Attack["⚔️ Attack Agents"]
        E1[SQLi Specialist]
        E2[XSS Specialist]
        E3[SSRF / IDOR]
        E4[Auth Bypass]
    end

    subgraph Validation["✅ Validation Agents"]
        F1[PoC Builder]
        F2[Impact Assessor]
        F3[Report Generator]
    end

    C --> Discovery
    C --> Attack
    Discovery --> |Findings| Attack
    Attack --> |Vulns| Validation

    style Orchestrator fill:#9b59b6,stroke:#8e44ad,color:#fff
    style Discovery fill:#3498db,stroke:#2980b9,color:#fff
    style Attack fill:#e74c3c,stroke:#c0392b,color:#fff
    style Validation fill:#2ecc71,stroke:#27ae60,color:#fff

🔄 Scan Execution Flow (click to expand)

sequenceDiagram
    participant U as User
    participant P as Phantom CLI
    participant O as Orchestrator Agent
    participant S as Sandbox (Docker)
    participant L as LLM Provider
    participant T as Target

    U->>P: phantom --target https://app.com
    P->>S: Launch sandbox container
    P->>O: Initialize with system prompt + tools

    Note over O,L: Reconnaissance Phase
    O->>L: Plan reconnaissance strategy
    L-->>O: Use nmap, nuclei, httpx
    O->>S: Execute: nmap -sV target
    S->>T: TCP/UDP probes
    T-->>S: Open ports & services
    S-->>O: Port scan results

    O->>S: Execute: nuclei -u target
    S->>T: Template-based checks
    T-->>S: Vulnerability matches
    S-->>O: Initial findings

    Note over O,L: Attack Phase
    O->>L: Analyze findings, plan attacks
    L-->>O: Spawn SQLi + XSS agents
    O->>S: Execute: sqlmap --url target/api
    S->>T: SQL injection payloads
    T-->>S: Database extracted
    S-->>O: Confirmed SQLi

    Note over O,L: Validation Phase
    O->>L: Build PoC for SQLi
    L-->>O: Working exploit script
    O->>S: Verify PoC execution
    S->>T: Reproduce attack
    T-->>S: Attack confirmed
    S-->>O: PoC validated

    O->>P: create_vulnerability_report()
    P->>U: 📋 Report with PoCs

🐳 Sandbox Architecture (click to expand)

graph TB
    subgraph Host["🖥️ Host Machine"]
        CLI[Phantom CLI]
        Docker[Docker Engine]
    end

    subgraph Sandbox["🐳 Phantom Sandbox Container"]
        ToolServer[Tool Server API]

        subgraph Tools["Security Tools"]
            nmap[nmap]
            nuclei[nuclei]
            sqlmap[sqlmap]
            ffuf[ffuf]
            httpx[httpx]
            katana[katana]
            semgrep[semgrep]
            arjun[arjun]
            zap[ZAP Proxy]
            more[+ 10 more...]
        end

        subgraph Runtime["Execution Environment"]
            Shell[Bash Shell]
            Python[Python Runtime]
            Browser[Playwright Browser]
        end
    end

    CLI --> |HTTP API| ToolServer
    ToolServer --> Tools
    ToolServer --> Runtime
    Sandbox -.-> |"Isolated Network"| Target["🎯 Target"]

    style Host fill:#2c3e50,stroke:#1a252f,color:#fff
    style Sandbox fill:#9b59b6,stroke:#8e44ad,color:#fff
    style Tools fill:#3498db,stroke:#2980b9,color:#fff
    style Runtime fill:#e74c3c,stroke:#c0392b,color:#fff

🧠 Knowledge & Memory System (click to expand)

graph TB
    subgraph "Persistent Knowledge"
        KS[(Knowledge Store<br/>JSON Persistence)]
        FP[False Positive<br/>Tracking]
        History[Scan History<br/>& Statistics]
    end

    subgraph "Working Memory"
        State[Agent State<br/>Messages & Context]
        Tracer[Telemetry Tracer<br/>Run Data]
        Graph[Attack Graph<br/>NetworkX DAG]
    end

    subgraph "Intelligence"
        MITRE[MITRE ATT&CK<br/>Enrichment]
        Compliance[Compliance<br/>Mapper]
        Priority[Priority Queue<br/>Severity-Ranked]
    end

    State --> |Findings| KS
    KS --> |Past Scans| State
    State --> |Vulns| Graph
    Graph --> |Attack Paths| MITRE
    MITRE --> |TTPs| Compliance
    KS --> |Known FPs| FP
    Tracer --> |Reports| History

    classDef persist fill:#f39c12,stroke:#e67e22,color:#000
    classDef working fill:#3498db,stroke:#2980b9,color:#fff
    classDef intel fill:#2ecc71,stroke:#27ae60,color:#fff

    class KS,FP,History persist
    class State,Tracer,Graph working
    class MITRE,Compliance,Priority intel

📊 Vulnerability Lifecycle (click to expand)

stateDiagram-v2
    [*] --> Detected: Tool finds potential vuln
    Detected --> Verified: PoC confirms exploitability
    Detected --> FalsePositive: Validation fails

    Verified --> Reported: create_vulnerability_report()
    Verified --> Exploited: Successful exploitation

    Exploited --> Reported: Impact documented

    Reported --> [*]: Scan complete

    FalsePositive --> [*]: Filtered out

    note right of Detected
        Nuclei, sqlmap, or
        manual testing finds
        a potential issue
    end note

    note right of Verified
        Agent builds PoC
        and confirms the
        vulnerability works
    end note

    note right of Reported
        CVSS scored, MITRE mapped,
        compliance tagged,
        remediation added
    end note

🚀 Quick Start

Prerequisites

• Docker (running) — Install Docker
• Python 3.12+ — Install Python
• An LLM API key — OpenAI, Anthropic, Groq (free), or any LiteLLM provider

Install & Run

# Install via pip
pip install phantom-agent

# Or via pipx (recommended for CLI tools)
pipx install phantom-agent

# Configure your LLM provider
export PHANTOM_LLM="openai/gpt-4o"        # or groq/llama-3.3-70b-versatile (free!)
export LLM_API_KEY="your-api-key"

# Launch your first scan
phantom scan --target https://your-app.com

Docker Quick Start

# Pull and run directly
docker run --rm -it \
  -e PHANTOM_LLM="openai/gpt-4o" \
  -e LLM_API_KEY="your-key" \
  -v /var/run/docker.sock:/var/run/docker.sock \
  redwan07/phantom:latest \
  scan --target https://your-app.com

[!NOTE] First run automatically pulls the sandbox image (ghcr.io/usta0x001/phantom-sandbox:latest, ~14GB, Kali-based with 20+ security tools). Scan results save to phantom_runs/.

📦 Usage

Basic Scans

# Web application black-box scan
phantom scan --target https://your-app.com

# Quick scan (faster, less thorough)
phantom scan --target https://your-app.com --scan-mode quick

# Deep scan (comprehensive, slower)
phantom scan --target https://your-app.com --scan-mode deep

# Non-interactive / headless mode (for scripts & CI)
phantom scan --target https://your-app.com --non-interactive

Advanced Testing

# Authenticated testing
phantom scan --target https://your-app.com \
  --instruction "Login with admin:password123 and test admin endpoints for IDOR"

# Focused vulnerability hunting
phantom scan --target https://api.your-app.com \
  --instruction "Focus on SQL injection and auth bypass in /api/v2 endpoints"

# Interactive TUI mode (rich terminal interface)
phantom --target https://your-app.com

Scan Modes

Mode	Speed	Coverage	Best For
quick	~10-20 min	Surface-level, common vulns	CI/CD gates, quick checks
standard	~30-60 min	Balanced depth	Regular security testing
deep	~1-3 hours	Full attack surface	Thorough pentests, audits

⚙️ Configuration

Environment Variables

Variable	Description	Example
PHANTOM_LLM	LLM provider/model	openai/gpt-4o
LLM_API_KEY	API key(s), comma-separated for rotation	sk-...
PHANTOM_REASONING_EFFORT	Thinking depth: low, medium, high	high
PHANTOM_SCAN_MODE	Default scan mode	standard
PERPLEXITY_API_KEY	Enable web search OSINT	pplx-...
PHANTOM_IMAGE	Custom sandbox Docker image	ghcr.io/usta0x001/phantom-sandbox:latest
PHANTOM_WEBHOOK_URL	Webhook for critical findings	https://hooks.slack.com/...
PHANTOM_DISABLE_BROWSER	Disable Playwright browser	false

Supported LLM Providers

Provider	Model Example	Notes
OpenAI	openai/gpt-4o	Best overall
Anthropic	anthropic/claude-sonnet-4-20250514	Excellent reasoning
Google	gemini/gemini-2.5-pro	Large context
Groq	groq/llama-3.3-70b-versatile	Free tier
Ollama	ollama/llama3.1	Local, no API key
DeepSeek	deepseek/deepseek-chat	Cost-effective
Azure	azure/gpt-4o	Enterprise
AWS Bedrock	bedrock/anthropic.claude-v2	Enterprise

Phantom uses LiteLLM — any supported provider works out of the box.

Persistent Configuration

# Save config so you don't re-enter it every time
phantom config set PHANTOM_LLM openai/gpt-4o
phantom config set LLM_API_KEY sk-your-key

# View current config
phantom config show

🔄 CI/CD Integration

GitHub Actions

name: Phantom Security Scan

on:
  pull_request:
    branches: [main]

jobs:
  security-scan:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4

      - name: Install Phantom
        run: pip install phantom-agent

      - name: Run Security Scan
        env:
          PHANTOM_LLM: ${{ secrets.PHANTOM_LLM }}
          LLM_API_KEY: ${{ secrets.LLM_API_KEY }}
        run: phantom scan --target ./ --non-interactive --scan-mode quick

      - name: Upload SARIF
        if: always()
        uses: github/codeql-action/upload-sarif@v3
        with:
          sarif_file: phantom_runs/latest/results.sarif

🔧 Development

# Clone the repository
git clone https://github.com/Usta0x001/Phantom.git
cd Phantom

# Create virtual environment
python -m venv .venv
source .venv/bin/activate  # Linux/Mac
# .venv\Scripts\activate   # Windows

# Install dependencies
pip install -e ".[dev]"

# Run tests
pytest tests/ -v

# Lint
ruff check phantom/

Project Structure

phantom/
├── phantom/                 # Core package
│   ├── agents/              # AI agent system
│   │   ├── PhantomAgent/    # Main agent + system prompt
│   │   ├── base_agent.py    # Agent base class
│   │   ├── state.py         # Thread-safe agent state
│   │   └── enhanced_state.py
│   ├── llm/                 # LLM integration
│   │   ├── llm.py           # Core LLM client
│   │   ├── provider_registry.py
│   │   └── memory_compressor.py
│   ├── core/                # Core modules
│   │   ├── scope_validator.py
│   │   ├── knowledge_store.py
│   │   ├── mitre_enrichment.py
│   │   ├── attack_graph.py
│   │   └── compliance_mapper.py
│   ├── tools/               # Security tool wrappers
│   ├── interface/           # CLI, TUI, reporting
│   ├── runtime/             # Docker sandbox management
│   ├── telemetry/           # Local-only run tracking
│   └── config/              # Configuration management
├── tests/                   # Test suite (90+ tests)
├── containers/              # Dockerfile for sandbox
├── scripts/                 # Install & build scripts
└── docs/                    # Documentation

📚 Documentation

Topic	Description
Quick Start	Get scanning in 2 minutes
Configuration	All settings and providers
Architecture	How the agent system works
CI/CD	GitHub Actions integration
Contributing	How to contribute

🤝 Contributing

We welcome contributions! See our Contributing Guide for details.

• 🐛 Bug Reports — Open an issue
• 💡 Feature Requests — Start a discussion
• 🔧 Pull Requests — Fork, branch, and submit a PR

⭐ Support

Love Phantom? Give us a ⭐ on GitHub!

📄 License

Apache License 2.0 — see LICENSE for details.

🙏 Acknowledgements

Phantom builds on incredible open-source projects: LiteLLM · Nuclei · Playwright · Textual · Rich · NetworkX

---

☠ PHANTOM · " Why So Serious ?! " · Autonomous Offensive Security Intelligence

Made with 🖤 by Usta0x001

[!WARNING] Only test systems you own or have explicit authorization to test. You are fully responsible for using Phantom ethically and legally. Unauthorized access to computer systems is illegal.