Skip to main content

Coaching Overview

Noxys provides in-context coaching to help users understand why a prompt was flagged and how to phrase it safely. Coaching is delivered through the browser extension as a non-blocking pre-send notification (Free / Pro tiers) or as a hard block with a justification flow (Enterprise tier with default-deny).

Goals​

  1. Reduce risk — convert risky prompts into safe ones before they reach the model
  2. Educate — turn each interaction into a learning moment, not a punishment
  3. Preserve productivity — never break the user's flow more than necessary

Coaching by Risk Signal Category​

Coaching content is organised by Risk Signal sub-category (see Glossary):

Each page contains:

  • Why it's risky — the threat model in plain language
  • Examples — anonymised real-world patterns
  • Best practice — how to rephrase or mask the prompt
  • Alternative phrasing — drop-in replacements
  • References — links to related policies, ADRs, regulatory frameworks

Tone & Style​

  • Empathic, never punitive — users are not adversaries
  • Action-oriented — every page ends with a concrete "Do this instead"
  • Bilingual EN/FR — French version available as <page>.fr.md
  • Versioned — content evolves with the threat landscape; see git history

Localisation​

The English version is canonical. French translations are maintained as sibling files (pii.md → pii.fr.md). When updating content, update both files in the same PR.

Note: This content is delivered to users at runtime via the extension's coaching panel and the console's policy editor. Keep it short, actionable, and free of jargon.