Security
Prompt Injection Defense for Agents That Read Untrusted Text
An agent that reads web pages and files can be hijacked by hidden instructions inside them. Here is the architecture that stops prompt injection.
least-privilege
Tags
Security
Least Privilege for CI Bots That Commit and Deploy Code
An automation bot with broad write and unlimited deploy rights turns one bad run into a whole-system failure. Here is how to shrink the blast radius.