llm-app-security
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill is a purely defensive educational resource. It provides technical guidance and code implementations for hardening AI applications against common attacks such as prompt injection and sensitive data exposure.
- [INDIRECT_PROMPT_INJECTION]: The skill describes and provides code for RAG (Retrieval-Augmented Generation) pipelines, which are a known attack surface for indirect prompt injection.
- Ingestion points: The
scan_documentandretrieve_for_userfunctions inSKILL.mdare designed to process external, potentially untrusted documents. - Boundary markers: The
build_messagesfunction implements strict message role separation and utilizes clear text delimiters (---) to isolate retrieved context from system instructions. - Capability inventory: The provided examples utilize file system access for document scanning, network requests for content moderation (targeting OpenAI's official API), and database operations for session management and vector retrieval.
- Sanitization: The skill explicitly implements multiple defensive layers, including malware scanning via ClamAV, PII redaction using Microsoft Presidio, and regex-based secret detection.
Audit Metadata