Adithyan Arun Kumar

A secure architecture for Hermes Agent and Mnemosyne with project-isolated recall, verified execution memory, safe writes, and rollback controls.

Run one autonomous AgentHound scan from a foothold, preserve the evidence in a local artifact, and turn it into an AI agent attack graph.

Trace untrusted data and poisoned instructions through AI agent context to high-impact tools and resources with AgentHound.

Detect suspicious MCP tool descriptions, map their reach to privileged capabilities, and inspect AgentHound's reversible ContextForge validation.

How Agentmask blocks secrets before they enter AI coding-agent context using hooks, redacted MCP tools, and layered detection.

Technical analysis of CVE-2025-45691, an arbitrary file-read flaw in RAGAS v0.2.3 through v0.2.14 caused by unsafe multimodal URL handling.

A technical walkthrough of diskless PowerShell shellcode execution using UnsafeNativeMethods, dynamic Win32 API resolution, and in-memory delegates.

A technical walkthrough of Windows sandbox-evasion primitives using FlsAlloc, delayed execution, Caesar encoding, and in-memory code execution.