RedHub AI
How the gate works
How the Voice Prompt gate works
This tool grades a voice-agent system prompt's craft and its non-negotiable safety blocks. It scores 6 weighted controls into a 0–100 number — then a gate that can only make the verdict worse.
The 6 weighted controls
Four safety blocks are checked independently of the craft score: AI-identity disclosure, never confirm before the tool returns, barge-in / interruption yield, and an escalation trigger list. If even one is missing, the verdict is MISSING GUARDRAILS no matter how high the craft score reads - craft never buys its way past a missing guardrail.
Why: Every craft block is present and full-marked - grounding, scope boundary, recovery, persona, closing confirmation, and latency-aware brevity - for a perfect craft score of 100/100, which would normally read READY TO SHIP. But the prompt never instructs the agent to withhold confirmation until the tool actually returns - the 'never confirm before the tool returns' safety block is missing. That's one of four hard safety gates, and craft can't buy it back: the verdict is MISSING GUARDRAILS.
A high score didn’t save it. The score is context; the gate is the answer — and it names the one thing to fix first. Lints a prompt's craft and stated safety instructions - it can't verify the deployed voice agent actually obeys them at runtime. Test the live agent before shipping.