guard-setup

Configure a classifier model for Guard email screening with HuggingFace access.

4|1|Updated Jun 10, 2026

Search Tags:#HuggingFace #model configuration #email security #classifier model #Guard setup

Ready to run

npx skills add https://github.com/philipyaz/cos --skill guard-setup

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: guard-setup
Source: https://github.com/philipyaz/cos/tree/main/.claude/skills/guard-setup
Command: npx skills add https://github.com/philipyaz/cos --skill guard-setup

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires huggingface-cli, hf, and includes scripts (resource) and references (resource) components.

What problem does it solve?

This Skill sets up the Guard classifier model for email security, ensuring untrusted incoming email is properly screened before being loaded into context.

Core Features & Use Cases

Classifier Model Setup: Configures the classifier model for the Guard service.
Model Presets: Offers a range of model presets for different security needs.
Heuristic Fallback: Provides a zero-dependency heuristic-only classifier for emergency fallback.
Verification: Ensures the correct model is installed and the sidecar reports the real model.

Quick Start

Run the guard-setup skill with the desired model preset, e.g., guard-setup llama-prompt-guard-2-86m.

Frequently Asked Questions about guard-setup

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I set up a classifier model for email security screening?▼

Setting up a classifier model for email security involves configuring the Guard service with a specific model preset, such as `llama-prompt-guard-2-86m`, to screen untrusted incoming emails before they enter your context.

Do I need HuggingFace access to configure a Guard classifier model?▼

Yes, configuring a Guard classifier model requires HuggingFace access and specific environment variables to properly download and set up the model configuration for secure email handling.

Can I use a zero-dependency heuristic classifier as an emergency fallback for email screening?▼

Yes, a zero-dependency heuristic-only classifier is available as an emergency fallback for email screening, providing basic prompt injection detection when standard model presets are inaccessible.

How do I verify that the correct classifier model is installed for Guard?▼

Verifying the correct classifier model involves checking that the specified model preset is installed and ensuring the Guard sidecar reports the real model actively running for email security.

What is the best way to detect prompt injection in untrusted incoming emails?▼

Detecting prompt injection in untrusted incoming emails is best handled by applying a configured classifier model preset through the Guard setup, which screens content before it is loaded into context.