mirror of
https://github.com/CyberSecurityUP/NeuroSploit.git
synced 2026-10-20 14:37:17 +02:00
CyberSecurityUP
48a40317f1
Replace the naive heuristic_select() with a recon-driven scoring engine that analyzes 30+ surface detection flags (WAF, CDN, SPA, frameworks, cloud, DB, auth, API, etc.) and uses them to: - Negatively filter agents for technologies NOT detected in recon (-10 penalty) - Conditionally apply baseline scores only when surface supports the class - Boost WAF-bypass/smuggling agents when WAF/CDN is detected (+8) - Reduce injection baseline behind WAFs (sqli_error gets +1 not +4) - Context-boost auth/API/SPA/CDN/template agents per detected surface - Expand signal mappings from ~23 to 60+ (frameworks, CDNs, WAFs, patterns) Also improve the LLM selector (SELECT_SYS) with explicit elimination rules and hard exclusions (no DB agents without direct DB port, no cloud agents without cloud endpoints), and double the recon context from 3000 to 6000 chars.