rowlab-labeling

Pass

Audited by Gen Agent Trust Hub on Nov 7, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest tabular data from external sources such as the web or S3 buckets and incorporate that data into prompts for the AI to answer specific questions per row. This creates a vulnerability where a row containing malicious instructions could influence the agent's behavior.\n
  • Ingestion points: Data is loaded from local disk, web URLs, or S3 using the dataset_get tool (referenced in SKILL.md).\n
  • Boundary markers: The instructions do not prescribe the use of structural delimiters to isolate external row content from the labeling instructions.\n
  • Capability inventory: The skill uses dataset_get for data retrieval and dataset_export for saving results, allowing for a read-write data flow.\n
  • Sanitization: The provided workflow does not include mechanisms for validating or sanitizing the input data before it is presented to the model.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 7, 2026, 11:53 AM