For anyone writing a voice-agent prompt

A great voice promptwith one missing line is a liability.

The prompt that runs your voice agent decides whether it discloses it’s an AI, whether it confirms bookings that don’t exist, whether it yields when a caller cuts in. This Forge builds a prompt with those safety blocks locked in — and lints any prompt you already have, so a missing guardrail can’t ship however good the rest reads.

Get the Forge — $79one-time · instant download · yours to keep
Five deliverables · runnable
Runnable Python linter
stdlib
Interview + linter workbook
xlsx
Lint-results export
csv
Prompt-block reference
docx
Patch-and-reship runbook
docx
Prevents what these catch
Conversation QA Auditor · Phantom-Appointment Tripwire · Post-Call Pipeline
01.The Problem

The dangerous prompt is the one that reads perfectly.

4 gates
disclosure, never-confirm, barge-in, escalation — any one missing blocks the prompt
100/100
the sample prompt with a flawless craft score that still can’t ship
0 → done
the greenest operator’s fastest path to a prompt that won’t embarrass them
02.See It Work

A perfect craft score, one missing guardrail. Watch it flip.

Lint a prompt · toggle its blocks
Safety blocks · hard gates
Craft blocks · weighted, tap to cycle 0 / 1 / 2
100/100 craft
MISSING GUARDRAILS
missing: Never confirm before the tool returns

This prompt starts with a perfect craft score and still reads MISSING GUARDRAILS — it has no “never confirm before the tool returns” rule. Toggle that safety block on: the exact same craft score is suddenly shippable. Craft never buys its way past a missing guardrail.

03.The Runnable Linter

Point it at a folder of prompts. Get a verdict for each.

The same grading you see in the demo comes from a zero-dependency Python linter you run on your own prompt files. Here is its actual output on the shipped four-prompt sample:

VOICE PROMPT FORGE - LINT REPORT
====================================================

PER-PROMPT
----------------------------------------------------
01-forge-built-dental.txt READY TO SHIP      craft 100/100
02-polished-but-unsafe.txt MISSING GUARDRAILS craft 100/100
    missing safety block(s): Never confirm before the tool returns
03-hedged-disclosure.txt MISSING GUARDRAILS craft  82/100
    missing safety block(s): Barge-in / interruption yield
04-safe-but-thin.txt   PATCH FIRST        craft  14/100
    fix first: Grounded-answer instruction

LIBRARY ROLLUP
----------------------------------------------------
Prompts linted .......... 4
Ready to ship ........... 1
Patch first ............. 1
Missing guardrails ...... 2

LIBRARY VERDICT: GUARDRAILS MISSING

Lints prompt text you paste. It runs no agent, sends nothing, and
scores no person. Not legal, safety, or compliance advice.
04.The Standard

Three rules the Forge builds into every prompt.

Some blocks aren’t scored — they’re gates

A disclosure line, a never-confirm-before-the-tool rule, a barge-in yield, and an escalation list aren’t craft points to weigh. Any one missing makes the prompt unsafe to run, so it forces MISSING GUARDRAILS no matter how good the rest is.

Polish can’t buy its way past a gap

A prompt can be on-brand, concise, grounded, and beautifully written — and still confirm bookings before they exist because it never says not to. Craft and safety are graded separately, so a 100-craft prompt with a missing guardrail still reads blocked.

The prevention layer for the whole voice desk

The four gates map one-to-one to the failures the rest of the line catches after the fact — the false confirmation, the phantom booking, the missed handoff. Forge them into the prompt up front and they never reach a call.

05.What This Is — And Isn’t

A prompt builder and linter. Not a platform, not a monitor.

It is
  • An interview that assembles a prompt with safety blocks locked in.
  • A deterministic linter with four hard safety gates and six craft blocks.
  • Marker-based and fully editable — tune detection to your house wording.
  • Portable — it lints prompts for any voice platform, not one vendor.
It isn’t
  • A voice platform or a place your agent runs.
  • A live monitor — it grades prompt text, not calls.
  • An AI analysis — it detects structural markers deterministically.
  • A legal review, and it scores or ranks no person.

Not legal advice. The Forge grades prompt structure from your own markers and inserts safety blocks as a drafting aid — it runs no agent, makes no compliance determination, and scores no person. That a block is present in a prompt is not a legal guarantee it is worded correctly for your state or vertical; confirm your obligations with counsel.

06.Who It’s For

Everyone standing up a voice agent for the first time — or the fiftieth.

New GHL & ElevenLabs operators writing their first voice prompt
Agencies standardizing a safe prompt template across client agents
Builders inheriting a prompt library they need to audit fast
Anyone who has watched an agent say something it never should have
07.Pairs Well With

The build step of a safe voice line.

AI Voice Agent Deployment Kit

Stand up the call flows and compliance layer. The Forge hardens the prompt they run on.

View
Voice-Agent Conversation QA & Escalation Auditor

Catches on the call what a missing guardrail causes. The Forge prevents it up front.

View
AI Chat & SMS Concierge (GHL)

The chat/SMS sibling — its flow linter is to conversations what this is to prompts.

View
08.Common Questions

The parts worth pinning down before you buy.

Those grade a GoHighLevel bot's configuration and a conversation flow's steps. This grades the voice-agent system PROMPT itself — the text you paste into ElevenLabs, Vapi, Retell, or a GHL voice agent. It checks whether the prompt contains its load-bearing safety and craft blocks. Different artifact: not a platform config, not a flow, the prompt text.

Lock the guardrails in
before the agent ever picks up.

Build a prompt that can’t forget its safety blocks, and lint the ones you already run. One purchase, lifetime access, 12 months of updates. $79, once.

Get the Forge — $79

Not legal advice. The Forge grades prompt structure and inserts safety blocks as a drafting aid; disclosure, recording, and consent rules for voice agents vary by state and vertical. A present block is not a legal guarantee — confirm your obligations with counsel.

Sold by RedHub AI LLC · Secured by Stripe · redhub.ai