An agent's guardrails only work inside the agent. The commands it leaves behind run later, in your terminal and your CI, with your keys. So I put the guardrail in the task runner, then tried to break it.
An agent's guardrails only work inside the agent. The commands it leaves behind run later, in your terminal and your CI, with your keys. So I put the guardrail in the task runner, ...
How I added proofreading, tone rewrites and in-context word meanings to my Chrome extension with the on-device Prompt API, and stopped the small model from replying to the text.
I wrote this post as part of the Kiroween 2026 hackathon. I am not a developer by training. I am...