A → LLM
Agent to LLM
- Virtual keys per agent
- Prompt guards and redaction
- Token and budget limits
Personal blog · security notes and projects
Notes and hands-on projects exploring how to make AI agents safer. I focus on how agents connect to AI models, tools, and one another and how to manage their access and keep track of what they do. I build small projects in my own time and share the code and lessons learned.
ABOUT ACK
ACK stands for Agentic Cyber Kit. It’s a place to share what I’m learning about AI security through practical projects. I publish the code, step-by-step labs, and notes so you can try them yourself.
coverage
Across the projects I work through the main areas of AI security, from the traffic between agents to the data they handle.
A → LLM
A → MCP
A → A2A
how attacks reach an agent through its inputs, and what stops them.
filtering what goes into and comes out of a model.
checking the tools an agent is allowed to load.
short-lived credentials instead of long-lived keys.
knowing which agent did what, and why.
keeping sensitive data out of prompts and logs.
projects
I let HolmesGPT investigate a Kubernetes incident with read-only access and give the repair to a separate operator agent with its own identity.
Read the project →notes
feedback
Found a mistake, have an idea, or want to talk about agentic security? Open an issue on GitHub or get in touch.