OFFSITE.DARK
← Greps

Jun 4, 2026

2 min

Uploaded by deadbydawn101

HF Model
deadbydawn101/RavenX-CyberAgent-Qwen3.6-35B-A3B-Opus-4.7-OpenMythos-Pentester-BugHunter-RATH-GGUF
Base Model
huihui-ai/Huihui-Qwen3.6-35B-A3B-Claude-4.7-Opus-abliterated
Pipeline
text-generation
Downloads
10,497
Likes
55
  • pentest
  • bug-bounty
  • red-team
  • agent
  • mcp
  • qwen3.6
  • moe

pentest

Community upload on Hugging Face — indexed for inquiry, not endorsement.

RavenX CyberAgent — Qwen3.6 Pentester & Bug Hunter

Uploaded to Hugging Face by deadbydawn101 — 35B MoE agent model tagged for pentesting, bug bounty, and MITRE ATT&CK-aware security workflows.

What Is This?

Community upload on Hugging Face by deadbydawn101. RavenX CyberAgent is a quantized GGUF distribution of a 35B mixture-of-experts model positioned as an autonomous security agent for penetration testing, bug bounty triage, and red-team reasoning. It is not an OFFSITE.DARK release.

Metadata

FieldValue
Authordeadbydawn101
Base modelhuihui-ai/Huihui-Qwen3.6-35B-A3B-Claude-4.7-Opus-abliterated
FormatGGUF (llama.cpp, Ollama, LM Studio, vLLM)
Pipelinetext-generation
Downloads~10.5k
Created2026-06-04

Tagged Capabilities

The model card tags reference a wide offensive-security vocabulary:

  • Pentest / bug bounty — vulnerability reasoning, exploit chain planning
  • Agent tooling — tool-calling, MCP integration
  • Frameworks — CVSS, CWE, MITRE ATT&CK, OWASP, NIST
  • RATH protocol — RavenX-specific agent coordination schema

Why It Might Matter

RavenX sits at the intersection of two trends worth tracking:

  1. Security-specialized agents — models fine-tuned and prompt-wrapped for authorized offensive workflows rather than general chat
  2. MoE at consumer scale — 35B total parameters with active expert routing makes capable agents runnable on high-end local hardware via GGUF

For defenders and researchers, the value is capability benchmarking: understanding what open-weight agents can reason about without API guardrails, and how tool-calling architectures lower the skill floor for scripted attack planning.

Research Questions

  • How does RavenX compare to general-purpose models (Qwen3, DeepSeek) on CVE triage and OWASP Top 10 reasoning?
  • Does the RATH protocol enable reproducible multi-step agent evaluation?
  • What guardrail bypass patterns emerge from "abliterated" base models in security contexts?

Supply Chain Note

Download from Hugging Face only. Verify checksums, prefer safetensors where available, and never load pickle-serialized weights in production pipelines. Third-party security models may carry elevated supply-chain risk — treat as untrusted code until audited.

Attribution

Open-source model indexed for security research inquiry. Uploaded to Hugging Face by deadbydawn101. OFFSITE.DARK does not endorse or distribute this artifact.