rag-injection-defense
Pass
Audited by Gen Agent Trust Hub on Sep 8, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill is a reference guide intended to teach developers how to handle untrusted data in RAG pipelines. It contains no executable scripts or system modifications.
- [PROMPT_INJECTION]: The content includes phrases like 'ignore previous instructions', but these are explicitly presented as examples of malicious payloads to be detected and discarded, rather than commands for the agent to follow.
- [INDIRECT_PROMPT_INJECTION]: The skill addresses the vulnerability surface where agents ingest untrusted content and offers sound security advice, including the use of quarantine zones for evidence and the preservation of safety postures regardless of retrieved content.
Audit Metadata