Coding Agent Security
Prompt injection · Secrets · Sandboxing · MCP · Skills · Detection
Defense-in-depth for teams using AI coding agents: treat everything the agent reads as an instruction, gate actions in code, sandbox and starve the network, keep secrets out of reach, vet MCP and skills like dependencies, and assume prevention fails.
Built for design decisions, day-to-day agent use, and reviews of AI-assisted development.
Securing Agentic AI
Prompt injection · MCP · Memory poisoning · Lateral movement · Threat modeling
Controls for AI agents that plan, call tools, keep memory, and talk to each other. Organized along five threat zones plus a cross-zone verification layer: the series takeaways on one page.
Built for design decisions, building and operating agent systems, and for architecture reviews.