AGENTGUARD · DESIGN & THREAT MODEL · 5 MIN READ
Why I put a proxy between my AI agent and my server
Why the enforcement point sits between an AI agent and an MCP server, and which failure modes require separate controls.
Writing
Engineering notes
Long-form notes from building and testing security boundaries for AI agents.
AGENTGUARD · IMPLEMENTATION & TRADEOFFS · 5 MIN READ
Rule-based prompt-injection detection: what it catches, what it costs
Why a detected injection blocks the whole result, why v1 uses deterministic rules, and what the tests found.