DrogonClaw

AI-Driven Offensive & Defensive Security Platform Official Website: drogonclaw.xyz

DrogonClaw is a next-generation cyber operations platform. Rather than acting as a simple wrapper for Kali tools, DrogonClaw operates as a Command-and-Control (C2) Brain. It understands objectives, plans attack workflows, adapts to new discoveries, and orchestrates a swarm of specialized autonomous agents through a unified intelligence core.

DrogonClaw focuses on high-confidence autonomous workflows, explainable findings, and reproducible evidence, avoiding the hallucinations common in early AI security tools.

Quick Start | Supported Providers | Architecture | Commands | Development | Setup Guide | Security | Disclaimer

Warning

Warning

Linux Only DrogonClaw is strictly designed and optimized for Linux-based operating systems (such as Kali Linux, Ubuntu, or Debian). It relies heavily on Linux-specific networking APIs, native filesystem permissions, and process management. It will not function on Windows or macOS.

What Works

• ReAct orchestration core — a hand-rolled Go intelligence loop (no Node.js/LangChain overhead) that plans, delegates, and self-corrects.
• Persistent intelligence graph — a JSON-backed memory store of typed entities (targets, assets, ports, services, vulnerabilities, credentials, flags) and their relationships.
• Sandboxed tool execution — commands run in isolated, ephemeral Docker environments with a host fallback that fails closed.
• Structured tool wrappers — typed wrappers for Nmap, Nuclei, Gobuster, FFUF, SQLMap, Subfinder, HTTPX, Checksec, Hydra, and forensics triage.
• Verified exploit templates — pre-compiled chains for known CVEs (EternalBlue, Log4Shell, PrintNightmare, MS08-067, Spring4Shell), validated against known-vulnerable configurations.
• Active Directory arsenal — impacket and BloodHound wrappers (Kerberoasting, Pass-the-Hash, DCSync) for domain pivoting.
• 7-state exploit parser — every exploit attempt is classified (e.g. SUCCESS_SHELL, PATCHED, FILTERED, WRONG_ARCH) and fed back to the reasoning engine.
• Dual-layer CVE intelligence — a rolling 120-day NVD cache plus an offline database of common CTF/pentest CVEs.
• Human-in-the-Loop safety — dangerous actions halt and require explicit operator approval.
• Remote C2 — an optional Telegram gateway for mobile command and a headless daemon mode.

Quick Start

Requirements

• Go 1.26+
• Docker (daemon running, for sandbox execution)

Install (From Source)

DrogonClaw operates outside of centralized registries to prevent censorship. It must be cloned directly from GitHub:

git clone https://github.com/0xP4X/drogonclaw.git
cd drogonclaw
go mod tidy
go build -o drogonclaw ./cmd/drogonclaw/

Once built, run ./drogonclaw to launch the terminal interface.

(Note: DrogonClaw is built with a Go control plane — it compiles directly to a native Linux binary.)

Install (From npm)

If you want the prebuilt Linux CLI from npm:

npm install -g drogonclaw
drogonclaw setup
drogonclaw

This package publishes the Linux x64 executable only. Non-Linux platforms are intentionally blocked at install time.

Configure

DrogonClaw is configured entirely through its Setup Wizard — there is no environment-variable or export-based configuration. On first launch the wizard walks you through everything and writes it to ~/.drogonclaw/config.json (owner read/write only):

1. Authorisation — scope/compliance acknowledgement.
2. Neural Provider — OpenRouter (default), NVIDIA NIM, OpenAI, Google Gemini, or local Ollama, plus model selection.
3. Credentials — the provider API key (or Ollama endpoint).
4. Remote C2 Gateway (optional) — Telegram bot token + chat ID.
5. Secondary API Keys (optional) — pick any of GitHub, Shodan, VirusTotal, Brave Search, Hunter.io, or Exa to enable OSINT/recon features. Skip the whole set, or choose only the ones you want.

Re-run it any time with drogonclaw setup or /setup inside the terminal.

./drogonclaw setup

Note: configuration is stored only in ~/.drogonclaw/config.json. DrogonClaw does not read provider/model settings from environment variables, so a stray export AI_PROVIDER=… in your shell can no longer override your saved setup.

Start

./drogonclaw

Run drogonclaw setup first if you have not configured a provider yet.

Daemon Mode (Headless)

Run the agent from your phone via the Telegram gateway without the terminal UI:

./drogonclaw daemon

Supported Providers

Provider	Config Key	Notes
OpenRouter	openrouter	Flexible multi-model gateway; model list is fetched live with a curated fallback
NVIDIA NIM	nvidia	High-performance inference (Nemotron, Qwen, DeepSeek, Llama)
OpenAI	openai	Direct API runtime (gpt-4o, gpt-4o-mini)
Google Gemini	gemini	Enterprise reasoning core (gemini-2.5-pro, gemini-2.5-flash)
Ollama	ollama	Autonomous offline runtime; point OLLAMA_BASE_URL at your local server

Provider credentials are configured via the Setup Wizard (./drogonclaw setup or /setup) and stored locally in ~/.drogonclaw/config.json (owner read/write only).

Architectural Pillars

flowchart TD
    subgraph UI ["User Interfaces"]
        CLI["CLI Terminal"]
        TG["Telegram C2 Gateway"]
    end

    subgraph Core ["DrogonClaw Engine (Go)"]
        Orchestrator["Custom Go ReAct Orchestrator"]
        Graph[("Intelligence Graph (JSON)")]
        LootDB[("LootDB (SQLite3)")]
        OPSEC["OPSEC Stealth Manager"]
        Swarm["Swarm Commander (Goroutines)"]
    end

    subgraph Execution ["Isolated Execution Environment"]
        Docker["Ephemeral Docker Sandbox"]
        Tools["Nmap, Metasploit, Custom Exploits"]
    end

    Target((("Target Network")))

    UI -->|Natural Language Instructions| Orchestrator
    Orchestrator <-->|Decoupled Memory| Graph
    Orchestrator -->|Delegates Parallel Missions| Swarm
    Orchestrator -->|Stealth Policy| OPSEC
    Orchestrator -->|Deploys Payloads| Docker
    Tools -->|Stores Loot & Credentials| LootDB
    Docker --- Tools
    Tools -->|Exploits & Recon| Target

Loading

The platform revolves around five major pillars:

1. The Orchestration Core

• Native Go ReAct Engine: Hand-rolled, hyper-fast intelligence loop free from heavy Node.js/LangChain overhead.
• Mission Planner: Breaks down objectives, reasons about paths, and delegates to specialized agents.
• Swarm Commander: Automatically spins up native OS threads (Goroutines) to execute concurrent mission vectors in isolated contexts.
• Intelligence Graph: A persistent JSON-backed memory graph that stores typed entities such as targets, assets, ports, services, vulnerabilities, credentials, and flags, plus explicit relationships between them.

2. The Intelligence & Exploit Ecosystem

A modular Go-backed tool registry allowing the agent to perform highly complex attacks without hallucinating syntax:

• Native Go Tool Wrappers: Structured wrappers for common pentest tools (Nmap, Nuclei, Gobuster, FFUF, SQLMap, Subfinder, HTTPX, Checksec, Hydra, Forensics Triage) that eliminate flag-guesswork and provide pre-optimised defaults.
• Verified Exploit Templates: Pre-compiled attack chains for critical vulnerabilities (EternalBlue, Log4Shell, PrintNightmare, MS08-067, Spring4Shell). Execution success depends on target vulnerability and environment; templates are verified against known vulnerable configurations, not guaranteed against all targets.
• Active Directory Arsenal: Native wrappers for impacket and BloodHound (Kerberoasting, Pass-the-Hash, DCSync) allowing the agent to pivot through Windows domains.
• 7-State Exploit Parser: The AI doesn't just read raw output. A deterministic parser classifies every exploit attempt into one of 7 states (e.g., SUCCESS_SHELL, PATCHED, FILTERED, WRONG_ARCH) and feeds the exact recommended next step back to the reasoning engine.
• Dual-Layer CVE Intelligence:
  • Dynamic local NVD cache indexing for the last 120 days of vulnerabilities.
  • Offline static database of the 100 most common CTF/pentest classic CVEs (vsftpd, dirtycow) for immediate recall.
• Binary Triage: Automated checksec, strings, nm, and gdb tools to feed binary context directly into the LLM context window.

3. User Interfaces

DrogonClaw is controlled through two primary interfaces:

• CLI Terminal: The interactive drogon> prompt with a stylized workspace, slash commands, and live operator feedback.
• Telegram C2 Gateway: A remote, mobile-friendly control channel that lets you issue natural-language instructions to the agent from your phone (see Telegram Gateway for setup and whitelisting).

4. The Cognitive Consciousness Loop

DrogonClaw no longer just blindly executes commands. It operates on a strict internal Cognitive Loop designed to catch logic flaws, correct its own mistakes, and act autonomously:

• PERCEIVE / REFLECT / CHALLENGE / ACT: Before every single tool execution, the AI explicitly evaluates the last output, hypothesizes why it failed, challenges its own assumptions (e.g., "Is there a 0-day here?"), and then acts.
• Dual-Persona Commander/Civilian: The AI acts as a ruthless Commander during execution, but switches to an inquisitive Civilian to pause and ask the operator for intuition when genuinely stuck via the ask_operator protocol.
• Zero-Day Hunting: Built-in fuzz_endpoint (ffuf integration) and analyze_source_code tools allow the agent to read raw source files and hunt for unpublished logic flaws, SQLi, and IDORs that no vulnerability scanner would ever catch.

5. Autonomous Execution & Safety Layer

DrogonClaw isolates operational risk and prevents unintended damage through:

• Human-in-the-Loop (HitL): Dangerous actions (like installing arbitrary software, modifying network interfaces, or dropping high-risk payloads) physically halt execution and prompt the operator with a full impact analysis before proceeding.
• Sandboxed Tool Execution: Running command-line tools in isolated, ephemeral Docker environments.
• Memory Failure Loops: The agent tracks failed command syntax globally; if an exploit syntax fails, the agent is mathematically prevented from repeating that exact mistake, forcing it to pivot.

Interactive Terminal & Commands

Inside the drogon> prompt, you can converse with the AI naturally or use specific slash commands:

• /setup - Launch the configuration wizard
• /skills - Show loaded module categories and usage guidance
• /skills <term> - Search modules by name, description, category, or parameter
• /skills <exact_name> - Inspect a module's required parameters and execution backend
• /status - Print runtime state and memory graph entity/link counts
• /health - Run sandbox/toolkit diagnostics
• /ctf <path> - Offline local-CTF triage with artifact inventory and verified flag detection
• /profile <target> - Build a passive, source-accountable profile before any focused research
• /auto - Toggle autopilot (skip Human-in-the-Loop approvals)
• /persona - Switch the agent persona
• /stealth - Adjust the OPSEC stealth posture
• /mode - Switch operational mode
• /report - Generate an engagement report
• /swarm - Manage the agent swarm
• /analyze <path> - Analyze a target artifact or source
• /sandbox - Inspect or toggle the sandbox runtime
• /queue - Show the pending task queue
• /copy - Copy the on-screen transcript to the clipboard (and save to drogonclaw_loot/drogonclaw_transcript.txt)
• F3 - Open the current output in your pager ($PAGER or less -R) so you can freely select and copy any part of it
• /help - Show the full command reference

Graceful Action Abortion: If DrogonClaw is running a long scan or executing an exploit and you want to steer it in a different direction, simply press Ctrl+C. This will instantly sever the active thread, halt all sandboxed executions, and drop you back to the prompt, preserving the session memory so you can inject new instructions.

Copying output: The terminal runs in the alternate screen, which has no scrollback, so you cannot select-and-copy arbitrary text from the UI (a drag highlights the whole screen). Press F3 to open the current output in your pager, where you can freely select and copy any part of it. Use /copy to export the full transcript to your clipboard and to ~/drogonclaw_loot/drogonclaw_transcript.txt.

Telegram Gateway

Allows you to text instructions to your agent from your phone. It runs automatically if you set TELEGRAM_TOKEN and TELEGRAM_CHAT_ID in your configuration (~/.drogonclaw/config.json).

Security Note: You must provide your TELEGRAM_CHAT_ID to whitelist your account, otherwise the agent will reject all commands.

Modularity & Swarm Intelligence

DrogonClaw is designed to scale into collaborative agent swarms. You can inject new specialized agents (e.g., a "Web Fuzzer Agent" or an "Active Directory Hound") without modifying the core orchestrator.

Development

DrogonClaw is a Go project. Use Go 1.26+ and have Docker available for sandbox execution.

make build        # build the binary for the current OS
make test         # run all Go tests with race detection
make lint         # run golangci-lint
make run          # build and launch the terminal interface
make daemon       # build and run in headless daemon mode
make docker-compose  # start core services via Docker Compose

Other useful targets: make test-cover, make format, make vet, make skills (regenerate the skill manifest), make doctor (system diagnostics), make clean.

Repository Structure

• cmd/drogonclaw/ - CLI entrypoint
• internal/ - engine, agents, tools, TUI, sandbox, memory, and domain packages
• docs/ - architecture and readiness documentation
• scripts/ - build, manifest, and asset generation scripts
• skills/ - executable module definitions
• supabase/ - backend/schema assets
• assets/ - logos and generated charts
• .github/ - repo automation and funding metadata

Security

DrogonClaw is built for authorized security testing only. Report suspected vulnerabilities in the project via SECURITY.md. Never deploy offensive capabilities against networks without explicit written consent.

Contributing

Contributions are welcome. For guidelines, the development workflow, and scope rules, see CONTRIBUTING.md. Ensure make build, make lint, and make test pass before submitting a pull request.

Disclaimer

DrogonClaw is designed for authorized security testing only. Always ensure you have explicit permission before testing any system. Unauthorized access to computer systems is illegal.

License

GNU AGPL v3

Star History

Note: GitHub restricted access to star data on 2026-06-30, so the live star-history.com embed no longer renders. This chart is a static SVG generated from the repo owner's star timestamps (no public token). Regenerate it with:

GITHUB_REPO=0xP4X/drogonclaw python3 scripts/star_history.py assets/star-history.svg

See https://star-history.com/blog/github-stargazer-api-restriction.