rag-architect
Pass
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill provides Python implementation templates for "Contextual/Agentic Chunking" and "Propositions-Based Chunking" (found in
references/chunking-strategies.md) that present a vulnerability surface for indirect prompt injection. - Ingestion points: The
documentandtextparameters in the provided code snippets are intended to ingest raw, potentially untrusted content from external files or knowledge bases. - Boundary markers: The prompt templates provided (e.g.,
"Provide a brief context for this document chunk... Document excerpt:\n\n{chunk}") do not include specific delimiters or "ignore embedded instructions" warnings to mitigate the risk of the LLM following instructions hidden within the processed data. - Capability inventory: The snippets utilize the OpenAI API client to process the ingested content, allowing the injected data to influence the LLM's logic.
- Sanitization: No sanitization, escaping, or structural validation is shown for the text data before it is interpolated into the prompts.
Audit Metadata