generate-rag-dataset
Pass
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill requires the agent to ingest and analyze external documents (PDFs, markdown, text files) and database schemas to generate question-and-answer pairs. This creates a surface where adversarial instructions embedded in those documents could potentially influence the agent's behavior during the analysis phase.
- Ingestion points: SKILL.md Step 1 directs the agent to read every document it can access in the knowledge base, including documents, schemas, and vector store configurations.
- Boundary markers: The instructions do not provide specific delimiters or guidance to the agent on how to isolate document content from potential instructions embedded within those documents.
- Capability inventory: The skill implies file-system read access (Step 1) and suggests file-system write access and Python code execution (Step 4) to export the final dataset.
- Sanitization: There are no instructions provided for sanitizing or filtering the content read from the knowledge base before it is processed by the agent.
- [COMMAND_EXECUTION]: The skill provides Python code snippets for exporting data using the pandas library and saving it to the local file system as a CSV file.
- Evidence: Step 4 includes a Python snippet:
import pandas as pd; df = pd.DataFrame(dataset); df.to_csv("rag_evaluation_dataset.csv", index=False). This is a standard operation for the stated purpose of the skill.
Audit Metadata