RedAmon v5.0.0 — AI Gauntlet

The biggest release yet. RedAmon now turns its offensive engine on AI and LLM systems themselves, alongside major recon, graph, and multi-agent upgrades.

🆕 AI Gauntlet — Offensive AI / LLM Testing

A full offensive testing suite for AI attack surfaces, integrating industry-standard tools:

• garak — broad single-shot LLM vulnerability scanner
• PyRIT — adversarial prompt testing with per-tool venv isolation
• Giskard — zero-egress LLM scanning
• promptfoo — eval-driven probing with unified reporting
• On-demand local LLM lifecycle (Ollama) with ref-counted, leased model management
• Target auth (3 modes) + custom off-graph targets
• Per-run timeout bounds and unified findings reporting

🔭 AI Surface Recon

• Adversarial AI surface recon across the full pipeline
• Endpoint AI classifier and JS Recon AI SDK detection
• AI Surface / Risk Red Zone tables

🕸️ Recon & Graph

• ZAP Ajax Spider resource enumeration (now active)
• Aggressive clustering pipeline for very large graphs
• IP-mode scope fixes, Naabu custom port handling, Endpoint.sources persistence
• JS Recon wall-clock timeout enforcement

🔥 Fireteam (Parallel Multi-Agent)

• Hardened wave-clock budgeting and per-member iteration limits
• Per-item exception handling, improved state management

🛠️ Fixes & Hardening

• GVM waits for Postgres healthy before starting gvmd (#115)
• Productivity loop: distinguishes debugging from stalling
• API key refresh on user settings, Bedrock content-shape fix

Full changelog: https://github.com/samugit83/redamon/compare/v4.0.0...v5.0.0