How does this skill prevent prompt injection?

It implements Layer 2 (Input Validation) with PII detection and Layer 5 (LLM Orchestration) which routes identifiers around the LLM rather than through it to prevent leakage.

Can I use this for multi-tenant applications?

Yes, it includes specific patterns for tenant isolation via immutable request contexts and query-level filtering to prevent cross-tenant data access.

Does it include logging and monitoring?

Yes, the Observability layer (Layer 8) provides patterns for sanitized logging, distributed tracing, and anomaly detection without exposing sensitive PII.

What is the 8-layer security architecture?

It is a comprehensive security framework covering Edge, Gateway, Input, Authorization, Data Access, LLM, Output, Storage, and Observability layers to protect AI applications.

Defense-in-Depth Security

Name: Defense-in-Depth Security
Author: yonatangross

byyonatangross

•

보안 및 테스팅

Implements a robust 8-layer security architecture to harden AI pipelines and protect LLM integrations from end-to-end vulnerabilities.

The Defense-in-Depth skill provides a comprehensive framework for securing AI applications by applying multiple validation layers from the network edge to data storage. It ensures that no single point of failure compromises the system, specifically addressing unique AI risks like prompt injection, cross-tenant data leaks, and PII exposure. By formalizing layers for input validation, immutable request contexts, and safe LLM orchestration, it enables developers to build production-ready, compliant, and highly secure AI-driven features within Claude Code.

주요 기능

01Multi-tenant isolation through mandatory query-level filtering

02Immutable request context for consistent and secure identity flow

038-layer security architecture spanning from Edge to Observability

04Standardized output validation and sanitized audit trail logging

05Secure LLM orchestration that prevents identifier leakage in prompts

0669 GitHub stars

사용 사례

01Establishing compliant audit trails and sanitized monitoring for sensitive AI pipelines

02Implementing strict multi-tenant isolation in RAG-based systems and vector databases

03Hardening production AI applications against prompt injection and unauthorized data access

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add yonatangross/orchestkit defense-in-depth

For use in Claude.ai and ChatGPT

Download Skill