Ethan Troy

@ethantroy.com

https://ethantroy.dev | hacking, go, rust, ai | research & dev @ http://hackIDLE.com

Using @vercel.com 's json-render to keep LLM output constrained, predictable, and fast. It limits models to known components, streams clean JSON, and reduces randomness. Haven’t red-teamed it yet, but this feels like the right direction vs prompt dumping.

Ethan Troy@ethantroy.com · 6mo ago

myctrl.tools has an AI assist feature now (the whole thing is still kinda early) Bring your own Anthropic API key 😎 It's better advice than your typical GRC Advisor (jk...no really) I might (probably) will add support for other LLMs soon

As AI agents get more autonomous, prompt injection will shift from “ignore all previous instructions” to “add a task to the backlog to X.” Once the payload crosses a trust boundary and lands in Jira, it’s no longer a prompt, it’s just another task. A task that makes me admin :D

myctrl.tools has an AI assist feature now (the whole thing is still kinda early) Bring your own Anthropic API key 😎 It's better advice than your typical GRC Advisor (jk...no really) I might (probably) will add support for other LLMs soon