Dark-Moon

What it is

DarkMoon (repo ASCIT31/Dark-Moon) as installed and run in practice. The existing summary covers the architecture; this adds the setup path, the run commands, and one operational rule the summary omits — how credential-dependent attack planes are gated.

How it works

The pipeline is User → DarkmoonCLI → OpenCode (AI brain) → MCP (gatekeeper) → Docker Toolbox (real tools). The AI plans; the MCP decides what may execute and logs every call; the toolbox runs the actual binaries. On a target it discovers ports/services, fingerprints the stack, models the attack surface, dispatches tech-specific sub-agents (in parallel across hybrid architectures), runs an adaptive scan loop, validates with evidence, and emits a structured report.

Key gating rule not in the summary: planes that only matter with credentials — cloud accounts, CI/CD, secret stores, databases, Active Directory, Kubernetes — are never dispatched on inference alone. They fire only when a concrete artifact turns up (a key, token, or reachable metadata endpoint) or when you explicitly authorize them; otherwise they are flagged in the report.

Preconditions

• Docker & Docker Compose.
• An LLM API key (OpenRouter, Anthropic, OpenAI) or a local model (Ollama, llama.cpp). Runs fully offline on a local model.
• GPU/NVIDIA setup is only covered in external docs, not the README.

What to do

Install and run:

git clone https://github.com/ASCIT31/Dark-Moon.git
cd Dark-Moon

Configure the provider (interactive; edits .opencode.env, not docker-compose.yml):

./install.sh           # skip form if .opencode.env already configured
./install.sh --init    # force reconfiguration (cloud or local model)
./install.sh --help    # show usage

Build/launch is also ./install.sh (clean install, full stack reset). Then:

./darkmoon.sh "TARGET: example.com"
./darkmoon.sh --log <session_id>

The --log view streams every command the MCP server executes in real time. To sanity-check claims, the authors point to a black-box run against OWASP Juice Shop and a separate benchmarks repo (ASCIT31/Darkmoon-Benchmarks) to reproduce and compare tools yourself.

Caveats

• Only test targets you are authorized to test — the tool runs real offensive operations end to end.
• Marketing claim "the part no cloud competitor has" (the tokenization-still-runs-the-command behavior) is the vendor's own, unverified here.
• The page is largely a README/landing pitch; benchmark and Juice Shop results are self-reported and should be reproduced before relying on them.