Building with agents 0/15

Building with agents

Agent safety

Prompt injection, leaked secrets and over-broad permissions, and the guardrails that limit what a misled or mistaken agent can do. An agent that reads untrusted text and can act on it needs these limits from the start.