llm-app-security
Pass
Audited by Gen Agent Trust Hub on Oct 26, 2026
Risk Level: SAFEPROMPT_INJECTIONINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATIONEXTERNAL_DOWNLOADS
Full Analysis
- [SAFE]: The skill is a documentation-oriented resource providing defensive security templates and best practices.\n- [PROMPT_INJECTION]: Includes extensive detection patterns and mitigation strategies to protect system instructions from override attempts.\n- [INDIRECT_PROMPT_INJECTION]: Documents risks associated with RAG pipelines and provides templates for secure instruction hierarchy and context delimitation.\n
- Ingestion points: build_messages function in SKILL.md; document retrieval in references/details.md.\n
- Boundary markers: Employs role separation and specific delimiters in message building templates.\n
- Capability inventory: Tooling examples include search_knowledge_base and create_ticket functions.\n
- Sanitization: Suggests delimiters and system rules to ignore instructions embedded in context docs.\n- [DATA_EXFILTRATION]: Provides implementation examples for PII scrubbing and secret detection to prevent sensitive data disclosure in LLM outputs.\n- [EXTERNAL_DOWNLOADS]: Documents integration with established third-party APIs and services for monitoring, rate limiting, and content moderation.
Audit Metadata